Embed Size (px)
Citation preview
1 23
Journal of Materials ScienceFull Set - Includes `Journal of MaterialsScience Letters' ISSN 0022-2461Volume 53Number 12 J Mater Sci (2018) 53:8720-8746DOI 10.1007/s10853-018-2134-6
Oxide-based RRAM materials forneuromorphic computing
XiaoLiang Hong, Desmond JiaJun Loy,Putu Andhita Dananjaya, Funan Tan,CheeMang Ng & WenSiang Lew
1 23
Your article is protected by copyright and
all rights are held exclusively by Springer
Science+Business Media, LLC, part of
Springer Nature. This e-offprint is for personal
use only and shall not be self-archived in
electronic repositories. If you wish to self-
archive your article, please use the accepted
manuscript version for posting on your own
website. You may further deposit the accepted
manuscript version in any repository,
provided it is only made publicly available 12
months after official publication or later and
provided acknowledgement is given to the
original source of publication and a link is
inserted to the published article on Springer's
website. The link must be accompanied by
the following text: "The final publication is
available at link.springer.com”.
REVIEW
Oxide-based RRAM materials for neuromorphic
computing
XiaoLiang Hong1, Desmond JiaJun Loy1, Putu Andhita Dananjaya1, Funan Tan1, CheeMang Ng2,and WenSiang Lew1,*
1Division of Physics and Applied Physics, School of Physical and Mathematical Sciences, Nanyang Technological University, 21
Nanyang Link, Singapore 637371, Singapore2School of Electrical and Electronic Engineering, Nanyang Technological University, 21 Nanyang Link, Singapore 637371,
Singapore
Received: 27 October 2017
Accepted: 10 February 2018
Published online:
20 February 2018
� Springer Science+Business
Media, LLC, part of Springer
Nature 2018
ABSTRACT
In this review, a comprehensive survey of different oxide-based resistive ran-
dom-access memories (RRAMs) for neuromorphic computing is provided. We
begin with the history of RRAM development, physical mechanism of con-
duction, fundamental of neuromorphic computing, followed by a review of a
variety of RRAM oxide materials (PCMO, HfOx, TaOx, TiOx, NiOx, etc.) with a
focus on their application for neuromorphic computing. Our goal is to give a
broad review of oxide-based RRAM materials that can be adapted to neuro-
morphic computing and to help further ongoing research in the field.
Introduction
Resistive random-access memory (RRAM) utilizes
the resistive switching (RS) phenomena to store
information, which offers new types of devices that
can outstrip the performance of traditional semicon-
ductor electronics devices. Compared to charge-
based memory devices, RRAM stands out due to its
smaller cell size 4F2, multi-bit capability as well as
energy per bit (* fJ/bit). The first report of resistive
switching phenomena is by Hickmott [1], where
resistive switching is found on SiOx, Al2O3, Ta2O5,
ZrO2 and TiO2. However, it was only after 38 years
that the heat of RRAM research was reawakened
with the observation of resistive switching behavior
in magnetoresistive films from University of Houston
[2]. Two years later, 64-bit RRAM array using Pr0.7Ca0.3MnO3 via a 500-nm complementary metal oxide
semiconductor (CMOS) process was reported by
Zhuang [3]. Between 2004 and 2007, Samsung and
Infineon [4] made a significant progress on RRAM
development with the first 3D RRAM array demon-
strated in 2007. The concept of RRAM used for neural
network and logic circuit was first published in Na-
ture by HP in 2008. The paper titled ‘‘The Missing
Memristor Found’’ [5] triggered another heat of
RRAM development. In the next 9 years, great
achievement had been witnessed from industries and
academies. The successful fabrication of 64-MB
RRAM test chip, 32-Gbit bilayer cross-point RRAM,
27-nm 16-Gbit Cu-based CBRRAM test chip and
4-layer 3D vertical self-selective RRAM array was
Address correspondence to E-mail: [email protected]
https://doi.org/10.1007/s10853-018-2134-6
J Mater Sci (2018) 53:8720–8746
Review
Author's personal copy
announced by Unity [6], SanDisk/Toshiba [7],
Micron/Sony [8] and IMECAS [9], respectively. In
2017, TSMC [10] announced its new plan to start
producing embedded RRAM chips in 2019 using a
22-nm process. Figure 1 shows the historical timeline
of important events on RRAM development in the
last half-century.
RRAM technology is compatible to the conven-
tional CMOS in a simple way [11], and the RRAM
embedded system could be integrated into IoTs,
automobile and infotainment platforms [12, 13]. In
addition to the memory hierarchy, RRAM has
demonstrated its application in the low-power com-
puting as non-volatile logic circuits and
neuromorphic computing as a synaptic device.
Recent reviews by H.S.P. Wong, Yu, S. M. and Aki-
naga, H [14–16] gave excellent and comprehensive
overviews of RRAM physical mechanism, materials,
performance and applications. Here, we limit this
review to the oxide-based RRAM materials where
neuromorphic computing applications have been
demonstrated. At first, fundamentals of RRAM and
neuromorphic computing will be discussed. There-
after, we will elaborate more on these oxide-based
RRAM devices (PCMO, HfOx, TaOx, TiOx, etc.) used
for neuromorphic computing application. This
review will end with a conclusion and outlook.
Figure 1 Historical timeline
of RRAM development from
1962 to 2017 [1–10].
J Mater Sci (2018) 53:8720–8746 8721
Author's personal copy
RRAM design and physical mechanism
A typical RRAM structure consists of three layers of
materials, namely (metal–insulator–metal) MIM
structure [17], which is a resistive oxide layer sand-
wiched between two electrodes. The resistance of the
device can be modulated by applying an external
electric field across the electrodes. RRAM devices
work based on chemical redox reactions, i.e., oxida-
tion and reduction reactions. In memory storage,
there are binary states, ‘‘0’’ and ‘‘1’’. ‘‘0’’ represents
data that is not stored, while ‘‘1’’ represents data that
is stored. In RRAMs, redox reactions form a filament
or bridge across the two metal layers, within the
insulator. When a filament is created, an electrically
conductive is formed between the two metal layers.
Hence, a low-resistance state (LRS) occurs when a
filament is formed, while a high-resistance state
(HRS) occurs when a filament is not formed or rup-
tured. When a filament is formed, it would be a ‘‘1’’
state where one bit of information is stored and when
a filament is ruptured, it would be a ‘‘0’’ state [18]
(Fig. 2). A typical hysteretic current–voltage (I–
V) characteristics in a metal–insulator–metal (MIM)
structure are shown in Fig. 3. The switching from
HRS to LRS is called the SET process, while the
switching from LRS to HRS is called RESET process.
In most situations, a forming voltage larger than the
SET voltage is needed for the pristine device to trig-
ger the resistive switching behaviors for the subse-
quent cycles. The SET and RESET processes can occur
at the same polarity of the applied voltage as shown
in Fig. 3a which is known as unipolar switching.
Similarly, bipolar switching refers to the switching
behaviors occurring at different polarities of voltage
(Fig. 3b).
The MIM stack of RRAM can be broadly classified
as symmetric and asymmetric structures. A sym-
metric MIM structure mainly exhibits unipolar
switching behavior, while a asymmetric structure
mainly exhibits bipolar switching behavior. The
mechanism of resistive switching behavior can
mainly be classified into the electrochemical metal-
lization effect (ECM) and the valence change memory
effect (VCM) [19]. ECM and VCM are typically
observed in RRAMs with one active metal electrode
(Cu, Ag, Ni) and one inert metal electrode (Pt, Ru, Au
or Ir). In ECM, the conductive path of the switching
layer is formed via metal cations of the electro-
chemically active metal electrode under an external
electric field. On the other hand, VCM works based
on the migration of anions where oxygen vacancies
contribute to the conductive path within the oxide
layer. In VCM, it typically requires an oxygen scav-
enging electrode to facilitate anionic movement
between the active and inert electrodes. ECM and
VCM are based on redox reactions of oxidation and
reduction. RRAM devices with switching layer
doped with Ti [20] or Ge [21] had been reported to
have a forming-free property where their initial
resistance state (IRS) is similar to the high-resistance
state (HRS). In addition, several other advantages of
doping in the RRAM fabrication process to improve
uniformity and frustration had been reported [22].
Besides the single switching layer, numerous inves-
tigations had been conducted on double and multiple
switching layers. It has been reported that a switch-
ing layer consisting of low-resistance and high-re-
sistance layers can reduce the randomness of resistive
switching. Additionally, multi-layer structures have
been promising for the multi-level storage function
[23, 24]. Table 1 summarizes the details of the RRAM
stack, electrodes and switching layers.
The conduction mechanism in oxide RRAM can be
analyzed by fitting the I–V characteristic of current
conduction in the HRS and the LRS. Various resistive
switching mechanisms have been proposed to
explain and model RRAM conduction behavior. They
include the Pool–Frenkel emission (P–F emission),
SCLC (spaced charge limited current), formation and
rupture of conductive filaments (CFs), electrode-
limited conduction (Table 2) [42]. Although the
underlying physical mechanisms are material/elec-
trode specific and not well understood, the
Figure 2 A schematic diagram of the mechanism of resistive
switching effect in RRAM cell, ‘‘1’’ state when a filament is
formed and ‘‘0’’ state when a filament is ruptured.
8722 J Mater Sci (2018) 53:8720–8746
Author's personal copy
conductive filament mechanism applies to a large
majority of binary oxide unipolar or bipolar RRAMs.
Neuromorphic computing
Comparing with traditional computational architec-
ture, the human brain working system has major
advantages in the following aspects. The energy
efficiency of the human brain (10 watts) [15] is
remarkably superior, reaching 5 orders of magnitude
lower than supercomputers. The human brain can
also adapt to different environments and is able to
perform complex processing. The advancement of
studies on the human brain system has led to a new
disruptive technology neuromorphic computing.
This technology imitates how the brain is working in
the design and implementation of the computational
system.
The ultrahigh energy efficiency of the human brain
is derived from its low-power neurons and spike-
based computation, which is mainly realized by four
components (Fig. 4). Neuron bodies function as an
integrator and a device for threshold spiking. In
order to implement a neuromorphic system, a
capacitor is used to mimic neuron bodies in neuro-
morphic architecture. Axons work as information
transmission connection which could be mimicked a
long wire. The signal input from multiple neurons to
a single neuron is provided by dendrite, a short wire
which plays the role of dendrite in the neuromorphic
system. The synapse is the most investigated com-
ponent that has been constructed so far, and it pro-
vides dynamical interconnections between neurons
by switching and plasticity. Memristor is the device
component that works as a substitution to a synapse.
In the neuroscience, synaptic weight of synapse refers
to the strength or amplitude of a connection between
two neurons which is mainly determined by the
amount of neurotransmitter released and absorbed.
The synaptic plasticity is the increase or decrease of
synaptic weight over the time. Short-term synaptic
plasticity acts on a timescale of tens of milliseconds to
a few minutes, while long-term plasticity lasts from
minutes to hours. Memristors are electrical devices
which can mimic the synapse. The resistance of
memristors, which functions as the synaptic weight
in the neuroscience, can be tuned as response to a
periodic voltage (or current) input, and the resistance
state can be retained when the power is turned off.
RRAM is one form of memristor, and its resistance
(synaptic weight) can be manipulated between high
and low resistive states by deliberately applied volt-
age, which causes the medium to acquire microscopic
conductive paths called filament. In addition, it is
possible to reproduce synaptic plasticity with RRAM
memory run by specific coding schemes. In the
human brain, 100 billion neurons are connected to
one another by 100 trillion synapses [43]. The neu-
romorphic computing processor consists of a vast
array of electronic spiking neurons and multi-bit
synapses realized using fully interconnected crossbar
memories.
There is a very large amount of fascinating work
performed to develop memristors for neuromorphic
systems. So far, several types of memristors have
been investigated including but not limited to RRAM
[44–49], phase change memory [50–61], spin device
[62–68], floating gate transistors [69–76] and an opti-
cal device [77–81]. Among them, RRAM is an excel-
lent memristor for non-volatile synapse application
as it enables higher density, scalability and efficient
Figure 3 Typical hysteretic
current–voltage (I–
V) characteristics in metal–
insulator–metal (MIM)
structures a unipolar switching
and b bipolar switching.
J Mater Sci (2018) 53:8720–8746 8723
Author's personal copy
chip design. When compared with other types of
synaptic device, RRAM possesses the advantages of
high endurance that FLASH memories do not offer,
lower programming energy and smaller cell size than
MRAM. In addition, the non-volatile property of
RRAM outweighs the traditional DRAM which
requires constant recharging [82–87]. Table 3 lists out
some of the leading institutions and industries using
Table 1 Summary of RRAM stack, electrodes and switching layer
Structures Switching
behavior
Model
RRAM stack
Symmetric
and
asymmetric
Symmetric structure:
(e.g., Pt/NiO/Pt [25], Pt/TiO2/Pt [26], Pt/
ZrO2/Pt [27], Pt/HfO2/Pt [28])
Mainly
unipolar
switching
Thermal dissolution model:
Oxygen ions accumulated near the anode during the SET.
The Joule heating activates O2- to combine with oxygen
vacancies during the RESET and rupture of conductive
filaments (CFs)
Asymmetric structure:
(e.g., Pt/TiO2/TiN [29], TiN/ZnO/Pt [30],
TiN/HfO2/Pt [31], Pt/NiO/Cu [32], Cu/
La2O3/Pt [33])
Mainly
bipolar
switching
Ionic migration model:
Oxide interfacial layer formation between electrode and
oxide acts as oxygen diffusion barrier.
Structures Mechanism Advantages
RRAM electrodes
Active and inert metal
electrode
Active metal electrode:
Ti, Al, Ag, Cu, Ni, TiN, TaN,
ZrNx, ITO
ECM
VCM
Mainly bipolar
switching
Active ? inert electrodes in RRAM exhibit better endurance
and longer retention time
Inert metal electrode:
Pt, Au, Ru, Pd
VCM
Mainly unipolar
switching
Novel electrodes:
CNT [34, 35] n ? Si [36]
graphene [37]
n ? -Si: self-rectifying property which is capable of
alleviating cross talk issues
Graphene and carbon nanotube electrodes: ultradense
memories.
Structures Mechanism Advantages
RRAM switching layer
Doping Dopants: Ti [20], Ge [21] Forming-free: Dopants diffuse into oxide layer
and oxygen vacancies are increased
Good data retention, potential multi-bit operation,
highly scalable property and fast switching
speed
Better uniformity (temporal fluctuations) (cycle
to cycle) and spatial fluctuations (device to
device)
Bilayer One layer with lower
resistivity and one with
higher resistivity
E.g., TMO1/TMO2 [24],
Ni/GeOx/HfOx/TaN [38],
MoOx/GdOx [39]
Resistive switching occurs at higher resistivity
layer, and the lower resistivity layer can
improve the resistance of the ON state
Low power consumption
Lower RESET current
Better uniformity (diffusion of metal ions from
the lower resistivity to the higher resistivity
layer. This stabilizes the CFs to reduce the
randomness of resistive switching (RS)
Multi-
layer
E.g., Pt/HfOx/TiOx/HfOx/
TiOx/TiN [40]., Pt/Ta2O5/
TiOxNy/TiN/Ta5/Pt [41]
Multi-level storage function
8724 J Mater Sci (2018) 53:8720–8746
Author's personal copy
resistive memories as synaptic devices. Phase change
memory (PCM) has shown great potential for neu-
romorphic computing in the Synapse project per-
formed by leading researchers from IBM (TrueNorth)
[88]. However, the greatest drawback is its high
power consumption. RRAM is touted as a possible
solution and has been studied by Nanoelectronics
and Nanotechnology Research Group from Stanford
University and NanoST laboratory from National
Chiao Tung University [89–92]. The results showed
that RRAM possesses similar performance as that of
PCRAM and consumes less power to operate artifi-
cial synapse [93–95]. Since then, RRAM has attracted
massive attention from numerous academic institu-
tions. From Fig. 5, the number of publications on
RRAM has increased by 14 times in the last decade.
Among these publications, the proportion of papers
demonstrating neuromorphic computing applica-
tions has increased every year since 2010. In 2016,
nearly 11% of papers presented RRAM’s applications
in the neuromorphic system. In general, most of
studies on RRAM involve oxide materials. Oxide-
based RRAM is attractive as the underlying metal
insulator–metal structure is simple, compact and
CMOS-compatible. Most importantly, the multi-level
behavior needed to imitate adaptive synaptic changes
Figure 4 Interconnectivity in neuronal circuit.
Table 3 Active players in the resistive memory of the synaptic devices (with to Ref. [97])
Project performer Synaptic device Device structure Scale Synaptic characteristics Application
GIST RRAM TiOx/HfOx Unit cell Multi-level cell Visual cortex
LETI CBRAM Ag/GeS2 8 9 8 1T1R N/A Human Cochlea and Retina
Stanford PCM GST 10 9 10 1T1R Yes Pattern recognition
POSTCH RRAM Pr1-xCaxMnO3 11 k Array Yes Signal recognition
IBM PCM Ge2Sb2Te5 165 K Array N/A Pattern recognition
Table 2 Conduction mechanism analysis
Mechanism Remarks
P-F emission (field-assisted
thermal ionization)
Thermal excitation of electrons emits from traps into the conduction band of the dielectric.
In(1/V) - V1/2
SCLC (space charge limited
current)
When electric field is high, the current is dominated by the charge carrier injected from the electrode.
The current is only dependent on the mobility rather than the charge-carrier density.
I–V2
Conductive filament Formation and rupture of conductive filaments
I–V
Not proportional to electrode area
Electrode limited Relevant to electrode (materials, area)
Schottky emission Thermal excitation of electrons emits over the barrier into the conduction band of the dielectric.
Direct tunneling Electron tunnels from cathode to anode directly
Thin film thickness less than 3 nm
Fowler–Nordheim (F-N) tunneling Electron tunnel from cathode to conduction band directly
High voltage
J Mater Sci (2018) 53:8720–8746 8725
Author's personal copy
has been found in many oxide-based RRAM devices.
In addition, the energy per synaptic operation can be
made as low as sub-pJ and the programming current
can reach below 1 lA [96]. In this review, we will
focus on oxide-based RRAM memristor device as a
synapse in the neuromorphic system.
The basis of learning and formation of memory are
the modification of synapse connections as a result of
accumulated experience, the feature known as
synaptic plasticity. So far, synaptic plasticity had
been widely studied their correlation in learning and
memory [98–100]. Hebbin’s rule postulates that
changes at synapses within the brain underlie learn-
ing and memory, which was proposed by Donald
Hebb in 1949 [101]. It provides an algorithm to
update weight of neuronal connection within neural
network and a physiology-based model to reproduce
the activity-dependent features of synaptic plasticity.
In biology, excitatory and inhibitory postsynaptic
potentials are transmitted between neurons through
chemical and electrical messaging at synapses, driv-
ing the generation of the new action potentials
(spikes). In RRAM-based neuromorphic approaches,
the tunable resistive state of RRAM functions as
synaptic weight in the neuromorphic system. The
weight of synapse could be manipulated by changing
the resistance states in RRAM devices, which follows
the rule of spike-timing-dependent plasticity (STDP)
[102–104]. STDP is a form of synaptic plasticity that
adjusts the strength of the connection between the
presynaptic neuron and postsynaptic neuron based
on the relative timing of its output and input action
potentials (electric pulses). In the implementation of
STDP learning rule, it requires the adjustment of the
connection strength between pre- and postsynaptic
neurons. The process of adjusting the strength of
these interneural connections is known as synaptic
operation. The number of synaptic operations of a
neuromorphic chip is strongly dependent on the
intrinsic properties of the materials used as the
synapse. This will further determine the lifetime of
the chip or its availability for learning process. The
required cyclability (number of synaptic operations)
of the chip will depend on the expected lifetime and
operation frequency of the chip. If the chip is expec-
ted to be available for learning for about 10 years
under 100 Hz of operation frequency, the chip must
be able to execute at least 1010 synaptic operations.
RRAM device has been reported to achieve endur-
ance of as high as 1012; thus, it emerges as one of the
strongest candidates for the development of highly
reliable neuromorphic chip. The feasibility of RRAM
device for synapse applications was first demon-
strated in 2010 by Jo et al. [53]. According to the
experimental RRAM STDP curve as shown in Fig. 6,
conductance increases with the time delay when the
presynaptic spikes precede the postsynaptic spikes
(t\ 0). Conversely, the decrease of conductance
occurs when the postsynaptic spikes precede the
presynaptic spikes (t[ 0). In addition to the rise/fall
of synaptic weight, the change of synaptic weight
could also be either long term or short term. Upon
changing the spike rate, Ohno [105] demonstrated
both the long-term potentiation and short-term
potentiation in Ag2S-based RRAM device. Long-term
potentiation refers to the persistent increase in
synaptic strength which lasts minutes or more, while
short-term potentiation refers to the transiently
Figure 5 Publications per
year from 2007 to 2016. The
data is from the web of science
site (www.webofknowledge.
com). The search expression
for blue bars is Topic = ‘‘R-
RAM’’ or ‘‘RRAM’’ or ‘‘Re-
sistive random-access
memory’’. The data for red
bars are from subsequent
refined with expressions
Refine = ‘‘neuromorphic’’ or
‘‘synapse’’.
8726 J Mater Sci (2018) 53:8720–8746
Author's personal copy
enhancement of synaptic strength which acts on a
timescale of tens of milliseconds to a few minutes.
The results showed that the RRAM device exhibits
short-term potentiation behavior with the applied
pulse width at 0.5 s and interpulse delay at 20 s. As
shown in Fig. 7, the potentiation of synaptic weights
is volatile and no permanent change was observed
after nine pulses. However, when the interpulse
delay was reduced by 10 times, 6 pulses are sufficient
to induce the stable potentiation and permanent high
conduction state remains thereafter. In general, short-
term and long-term plasticity could be mimicked in
the single synapse by tuning the non-overlapping
spike rate. This synaptic plasticity plays a crucial role
in the way the brain implements learning and
memory. Therefore, the rule of STDP (Fig. 7) is con-
sidered as one implementation of Hebbian learning.
Oxide-based RRAM for neuromorphiccomputing application
Resistive switching behavior had been observed in
many oxides, while most of them are transition
metals. Some of them had been explored for neuro-
morphic application and summarized in the periodic
table (Fig. 8). We selected PMCO, HfOx, TaOx, TiOx,
NiOx, AlOx, WOx and other oxides for discussion in
this article as their property and application for
neuromorphic computing were well investigated.
The growth techniques of RRAM oxide materials
include traditional deposition methods, i.e., sputter-
ing, atomic layer deposition (ALD) and pulsed laser
deposition (PLD). In addition, novel methods for
RRAM fabrication have also been reported. The tra-
ditional and novel methods for RRAM are listed in
Table 4.
Pr12xCaxMnO3 (PCMO)-based synapticdevices
Pr1-xCaxMnO3 (PCMO), as representative materials
in perovskite transition metal oxide, has been widely
employed as synaptic devices. Its resistive switching
property was first discovered by Asamitsu et al. in
late 1990 [135], which draws the attention from aca-
demies on complex transition metal oxide for RRAM
device [136, 137]. The resistive switching of PCMO is
based on metal–insulator (MIT) mechanism. MIT is a
kind of valence change system, in which the electrons
are injected into the insulator layer to conduct the
current with an external applied field. In PCMO, the
injected electronic charge distorts the superlattice
structure and the mixed valence band behavior, a
Figure 6 Experimental RRAM STDP curve. (Reprinted with
permission from Ref. [53].).
Figure 7 a Short-term
potentiation and b long-term
potentiation of Ag2S-based
devices are induced for
different spike rates.
(Reprinted with permission
from Ref [105].).
J Mater Sci (2018) 53:8720–8746 8727
Author's personal copy
process similar to ion doping processes. In addition,
PCMO is LRS active materials in which the SET
operation is associated with reverse biasing, while
RESET operation occurs at forward biasing.
In 2012 and 2013, Park [138, 139] demonstrated
1-kbit cross-point array neuromorphic system based
on PCMO synaptic devices. The HRS is achieved
with the applied positive bias when the oxygen ions
are attracted from the PCMO layer, thus forming a
thicker oxide layer with Al electrode. The continu-
ously increasing potentiation and depression behav-
iors were found when the identical spikes (pulses
with same amplitude and width) were applied. In
order to overcome the difficulty in achieving gradual
long-term depression (LTD) caused by asymmetry
potentiation/depression, a new programming
scheme was developed [140]. The growth process
control of PCMO device with TiN electrode was
studied to optimize the device electric performance
[141]. Similarly, to avoid an abrupt LTD, a two-
PCMO-memristor device model was proposed by
Moon et al. [142]. In 2015, a circuit capable of realiz-
ing the learning process was designed and published
[143]. In the same year, a high-density PCMO cross-
point synapse array on the 8-inch wafer was fabri-
cated. The proposed system had been demonstrated
to recognize human thought patterns for three vow-
els [97]. In addition, vivo experiment had also been
performed by the group. In the paper, the PCMO-
based nanoscale analogy synapse device was suc-
cessfully applied for neural fear-conditioning signal
recognition on a live rat. In 2016, Moon, K from
Hwang, H group reported PCMO-based interface
switching device with 5-b MLC (32 levels) and
improved data retention by using Mo electrodes for
the neuromorphic system [144]. The Mo/PCMO
Figure 8 The periodic table of the elements highlighted with the host metal of RRAM materials and those had demonstrated for
neuromorphic computing.
Table 4 Summary of RRAM
device fabrication methods RRAM device deposition methods
Traditional methods Sputtering [106–113]
Atomic layer deposition [23, 114–117]
Pulsed laser deposition [118–122]
Plasma enhanced chemical vapor deposition [123–125]
Chemical synthesis particle ? spin-coating [126–128]
Sol–gel [129–132]
Novel methods Shattering process using an atomic force microscopy (AFM) tip [133]
Semi-auto screen printer [134]
8728 J Mater Sci (2018) 53:8720–8746
Author's personal copy
synapse array has experimentally confirmed with the
realization of pattern recognition with high accuracy
when working with NbO2 oscillator neuron [145]. In
2017, a comprehensive study of PCMO-based
synaptic devices had been reported in the book
‘‘Neuro-inspired Computing Using Resistive Synap-
tic Devices’’ [146]. The PCMO synapse performance is
evaluated in 1-kb and 8-kb crossbar arrays. The
interfacial-type PCMO RRAM was demonstrated
with wide on/off ratio (104), extremely stable analogy
resistance changes and low LRS current (1 lA) when
device scales down to 150 nm. The influence of active
electrode, nitrogen treatment and extra insulating
layer (AlOx) has also been investigated. The synaptic
performance of PCMO is evaluated as shown in
Fig. 9, where the potentiation and depression with
identical and non-identical spikes are displayed. It is
observed that the gradual switching of depression is
only triggered by non-identical spikes, which is
accounted for the high asymmetry ratio of potentia-
tion/depression conduction.
So far, PCMO is one of the most matured RRAM
materials for neuromorphic computing. It possesses
various advantages: nanoscale device dimension
(150 nm), multi-level states (5-b MLC), retention,
uniformity (STD(r)/Iave\ 0.5), on/off ratio (104) and
high-density crossbar array structure (8 kb) [146].
However, further improvement of symmetric weight
update is still required.
HfOx-based synaptic devices
HfOx materials have been widely used in RRAM
materials because of its excellent CMOS compatibil-
ity. In industries, HfOx has been employed as a high-
k dielectric for the gate insulator of CMOS MOSFETs.
In HfOx-based RRAM device, TiN is usually applied
as an electrode. TiN functions as an oxygen scav-
enger that depletes the O atom from HfOx thin film
and functions as an oxygen reservoir. It has been
reported that HfOx RRAM has a large on/off ratio
([ 102), high endurance ([ 106), retention, multi-bit
storage and high-speed operation (\ 10 ns) [142–146].
In 2011, a novel electronic synapse of HfOx/AlOx-
based memory was presented by Yu et al. The
structure showed attractive features such as reduced
randomness of resistive switching, multi-level
switching and the capability to modulate resistances
based on the input pulse amplitudes, suggesting
great potential to use in emerging neuromorphic
computation system [147]. In 2012, the same group
reported HfOx-based RRAM device application in the
neuromorphic visual system. At the system level, a
neuromorphic visual system consisting of 1024
CMOS neuron circuits and 16,348 RRAM synaptic
devices was fabricated. The system exhibited great
performance in image orientation selectivity and high
tolerance on temporal and spatial variability. After
one year, Gao, B [148] and his coworkers developed a
method to achieve gradual switching on the setting
process in TiN/HfOx/Pt devices. To avoid the gen-
eration of random and avalanching oxygen vacancies
Figure 9 Potentiation and depression with various programming spikes: identical and non-identical spikes, reprinted with permission
from Ref. [146].
J Mater Sci (2018) 53:8720–8746 8729
Author's personal copy
(VO) clusters, the team developed a methodology by
doping trivalent elements into HfOx layer. The local
formation energy of VO near the dopants was
reduced, and VO will distribute more uniformly in
the conductive filament region. The fabricated TiN/
Gd:HfOx/Pt devices have also shown controllable
multilevel resistive switching behavior. Garbin [149]
applied pulse-train schemes to a 3 bit per cell HfO2
RRAM in 2014, causing the relative standard devia-
tions of resistance levels to improve up to 80% com-
pared to the single-pulse scheme. In the same year,
Benoist, A integrated TiN/HfO2/Ti/TiN RRAM with
an advanced 28-nm CMOS process [150]. A global
overview of HfO2 material performances was asses-
sed on a statistical basis, and projection for larger
array integration was discussed. The influence of Cu
dopant was investigated by Tingting, G., and it was
found that Cu-doped HfO2-based RRAM has
improved resistive switching with multilevel storage
[151]. The application of Mn-doped HfO2-based
RRAM has demonstrated speech recognition by
Mandal [3]. A comparison between a 20-nm times
20-nm sized synaptic memory device with that of a
state-of-the-art VLSI SRAM synapse has exhibited 103
reduction in area and 106 times reduction in the
power consumption per learning cycle. The colormap
for speech recognition of words ‘‘Hello’’ and ‘‘Apple’’
is shown in Fig. 10.
In 2015, Wang presented a new artificial synapse
scheme, consisting of a HfOx RRAM memristive
switches connected to 2 transistors responsible for
gating communication and learning operations.
STDP was achieved through appropriate shaping of
the presynaptic and the postsynaptic spikes [152].
Covi, E demonstrated the Al:HfO2 memristor device
use for artificial synapse, emulating the potentiation
and depression processes using an easily imple-
mentable algorithm based on a train of identical
pulses [153]. The impact of HfOx-based synapses
cycle to cycle (temporal) and device to device (spa-
tial) variability on the convolutional neural network
was investigated by Garbin, D [154]. The results
showed that the convolutional neural network (CNN)
architecture has a high tolerance to the variability.
Lee et al. [155] proposed a silicon-based charge trap
memory with Al/HfO2/Al2O3/Si3N4/Si structure.
The device implemented both short-term plasticity
and long-term potentiation in the synapse. The use of
HfO2-based oxide-based resistive memory (OxRAM)
devices operated in a binary mode to implement
synapses in a CNN was also studied by Garbin, D in
2015 [149]. The proposed HfO2-based OxRAM tech-
nology offers good electrical properties including
high endurance ([ 108 cycles), fast speed (\ 10 ns)
and low energy (\ 10 pJ). High accuracy (recognition
rate[ 98%) was demonstrated for a complex visual
pattern recognition application. Covi, E proposed
TiN/HfO2/Ti/TiN memristor as artificial synapse
for neuromorphic architectures. The collected STDP
data were used to simulate a simple fully connected
spiking neural network (SNN) for pattern recognition
[156]. An innovative approach for real-time decoding
of brain signals based on SNN was presented by
Werner [157]. HfOX-based RRAM devices were used
to implement synapse and SNN, enabling the net-
work for autonomous online spike sorting of mea-
sured biological signals. The system allows real-time
learning and completely unsupervised operation. The
real-time functionality, low power consumption (10
nW) and high recognition rate 90% of the system
make it a great candidate for future healthcare
applications. HfOx-based synapses were used in SNN
Figure 10 Synaptic weights. a Initial current level. b The con-
ductance distribution when the word ‘‘Hello’’ trained on the
crossbar array of synaptic devices. c) The conductance distribution
when the word ‘‘Apple’’ trained on the crossbar array of synaptic
devices. Reprinted with permission from Ref. [3].
8730 J Mater Sci (2018) 53:8720–8746
Author's personal copy
for real-time unsupervised spike sorting of complex
brain signals. Synaptic weights are modulated
through the application of an online learning strategy
inspired by biological STDP. According to the report
by Wang et al. [158] this network had been tested by
real spiking data from the in vitro preparation of the
crayfish sensory-motor system. The proposed RRAM
was validated by different sets of real biological
spiking data without parameter tuning (e.g., the
threshold level for spike detection). This artificial
SNN is able to identify, learn, recognize and distin-
guish between different spike shapes in the input
signal with a recognition rate about 90% without any
supervision.
HfOx is a superior RRAM material and the HfOx-
based synaptic device has demonstrated its applica-
tion in neuromorphic computing, such as speech or
pattern recognition. Gradual switching in the SET
process can also be achieved through doping with
trivalent metals, enabling the high-density
applications.
TaOx-based synaptic devices
TaOx-based synaptic devices have also attracted
much attention. TaOx switching layer in an RRAM
device usually consists of two phases: TaO2 phase
which is more conducting and Ta2O5 phase which is
more insulating. The TaOx RRAM has been reported
to possess high endurance ([ 1012 cycles), and its
application in neuromorphic computing has been
well demonstrated. TaOx could be engineered in
synaptic device alone or combined with another
oxide (TiO2) layer in MIM structures. In this section,
both single-layer TaOx or multi-layer TaOx-based
synaptic devices will be discussed.
Wang reported a high-density 3D synaptic archi-
tecture based on Ta/TaOx/TiO2/Ti RRAM with
ultralow sub-10 fJ energy per spike for neuromorphic
computation in 2014 [159]. In addition, the analogue
synaptic plasticity was simulated using the physical
and compact models to facilitate future neuromor-
phic system designs. Thereafter, Wang et al. further
investigated the synaptic device with the same
structure experimentally and theoretically. The
results showed that the device can be used for
implementing concurrent inhibitory and excitatory
synapses (Fig. 11). In addition, the devices also pro-
vided superior performances as compared to typical
filamentary synaptic devices in reducing synaptic
conductance and state fluctuation [160]. In the report,
a physics-based compact model was also proposed
for facilitating circuit-level design. Based on previous
studies, the group successfully applied 3D two-lay-
ered Ta/TaOx/TiO2/Ti cross-point synaptic array to
implement highly anticipated hardware neural net-
works (HNN) in 2016 [161]. In total, more than 50
analogue synaptic weight states were controlled with
minimal drifting during a cycling endurance test of
5000 training pulses. The team also proposed a new
state-independent bipolar-pulse-training (BP)
scheme, and this scheme significantly improved the
nonlinearity of weight updates, consequently
improving the training accuracy.
Similarly, Ta/TaOx/TiO2/Ti synaptic devices were
also reported by Gao et al. [162]. The device is
forming-free, and more than 200 levels of conduc-
tance states could be continuously tuned by identical
programming pulses. In addition, the team proposed
a novel fully parallel write and read scheme to
accelerate the weight update and increase energy
efficiency in the training process on the chip. In order
to engineer the conduction modulation linearity,
Wang et al. [163] proposed a new approach in 2016
where a diffusion limiting layer (SiO2) is inserted at
the TiN/TaOx interface. The device exhibited higher
switching linearity and lower power consumption
that is desirable in neuromorphic computing hard-
ware. Recently, Pt/Ta2O5-x/W devices fabricated
using vertical 3D architectures were demonstrated by
Wang [164]. The proposed memory can be recovered
with a timescale when the electrical stimulation was
removed. This recoverable process can emulate the
synaptic plasticity including rapid decay and slow
decay stages of forgetting in the memory loss process
of the human brain. Yao, P and his coworkers
reported face classification using electronic synapses
of TaOx/HfAlyOx-based RRAM devices [165]. This
optimized device structure has shown bidirectional
continuous weight modulation behavior. A neuro-
morphic network is developed using analogue
1024-cell-1T1R RRAM array, and its application in
grayscale face classification has been experimentally
demonstrated with parallel online training. The
energy consumption in each iteration is
1000 9 (20 9) smaller than the Intel Xeon Phi pro-
cessor with off-chip memory.
From our literature review, bilayer stack TaOx/
(HfAlyOx, TiOx) is intentionally designed as a
synaptic device in the application of neuromorphic
J Mater Sci (2018) 53:8720–8746 8731
Author's personal copy
computing. The possibility of engineering the TaOx-
based RRAM array in 3D structure enables the
emulation of the high-density synaptic network in the
real biological system.
TiOx-based synaptic devices
TiOx is one of earliest materials found to exhibit
resistive switching property. In the previous section,
TaOx/TiOx bilayer structure RRAM for synaptic
application has been introduced. Here, we will dis-
cuss other multi- and single-layer TiOx-based
synaptic devices.
In 2011, the synaptic performance of TiO was
explored by Seo [46], demonstrating that TiOx and
TiOy bilayer RRAMs exhibit analogue property
where the multilevel conductance states were caused
by the movement of oxygen ions between the two
TiOx phase. The STDP was demonstrated by apply-
ing 100 successive identical pulses for potentiation or
depression of synaptic weight. A year later, Yu
developed TiOx/HfOx/TiOx/HfOx multi-layer
RRAM stacks exhibiting more gradual and smooth
RESET switching for synaptic devices [166]. Short
pulses (10 ns) with an identical pulse amplitude
enabled a sub-pJ energy per spike with potentially
simple neuron circuits. Berdan emulated short-term
synaptic dynamics with TiO2 memristive devices in
2015 [167]. The meta-stable memory transitions in
TiO2 RRAM devices have proved to be the key fea-
ture to capture short-term synaptic dynamics. In
addition to Seo, K works on bilayer TiOx/TiOy
RRAM structure in 2011 [46], Bousoulas, P studied
amorphous–crystalline interfaces in TiOx/TiOy
RRAM structures for enhanced resistive switching
and synaptic properties [168]. A physical model was
also proposed to divulge the crucial role of temper-
ature, electric potential and oxygen vacancy density
on the switching effect. An 8 9 8 array of the neuron
on a standard 6 M 180 nm of CMOS process as part
of a larger multi-purpose neuromorphic chip was
demonstrated by Mostafa, H as shown in Fig. 12
[169]. The proposed CMOS-memristor system com-
prises of CMOS neurons interconnected through TiOx
memristors and spike-based learning circuits which
modulate the conductance of the memristive synapse
elements according to a spike-based perceptron
plasticity rule. In 2016, Park developed a Mo/TiOx-
based interface RRAM and proposed a hybrid pulse
mode for the synaptic application [170]. The TiOx-
based device is capable of producing 64-level con-
ductance states, and the proposed hybrid pulse mode
improves the symmetry of conductance change
under both potentiation and depression conditions.
The neural network simulation has shown that the
hybrid pulse mode can enhance the pattern recogni-
tion accuracy of the proposed TiOx stack.
Although extensive investigations have been car-
ried out on switching mechanism of TiOx RRAM, the
demonstration of TiOx RRAM for neuromorphic
computing is still limited when compared to the
above-mentioned PCMO, HfOx, TaOx materials. In
most situations, TiOx is used in conjunction with
bilayer or multi-layer synaptic device to optimize the
performance.
Figure 11 Spike-timing-dependent plasticity (STDP). a Biomor-
phic-action-potential-like waveforms used for STDP measurement
and simulation. The prespike pulse was applied to the Ta electrode,
and the postspike pulse was applied to the Ti electrode. The
potentiation and depression actions were controlled by the relative
timing of the pre- and postspikes (Dt). The synapse weight change
(Dw) was positive for Dt[ 0, whereas it was negative for Dt\ 0.
b The measured Dw as a function of Dt for the Ta/TaOx/TiO2/Ti
device. The red line shows the fitting result of using the physics-
based compact model. c Simulation results for STDP, obtained by
considering oxygen ion migration and the homogeneous barrier
modulation model. Reprinted with permission from Ref [160].
8732 J Mater Sci (2018) 53:8720–8746
Author's personal copy
NiOx-based synaptic devices
Similar to TiOx, NiOx is one of earliest materials
found to exhibit restive switching behavior. Although
NiOx-based RRAM devices have been reported with
high endurance (106) and retention, its application for
neuromorphic computing is restricted due to the
poor uniformity. A bipolar NiOx RRAM, two NMOS
FET and a capacitor were used to fabricate analogue
synaptic devices (circuits) by Akoh [171]. This device
also has the ability to update the synaptic conduc-
tance according to the difference of pre- and post-
neuron spike timing. In 2013, Hu emulated the
paired-pulse facilitation of biological synapse with
NiOx-based memristor, thus creating a form of short-
term synaptic plasticity [172]. It was also observed
that the current in NiOx-based memristor had
increased after a second electrical pulse. Addition-
ally, the magnitude of the facilitation decreases with
the pulse interval and it increases with the pulse
magnitude or pulse width. Hu, S.G and his cowork-
ers realized the well-known Pavlov’s dog model with
NiOx-based memristor [173]. The long-term
potentiation behavior was emulated by the increase
in memristor’s conductance when applying the elec-
trical pulses. In addition, spontaneous conductance
decaying toward its initial state resembles the
synaptic long-term potentiation. Finally, an artificial
neural network was constructed to realize the Pav-
lov’s dog model (Fig. 13).
WOx-based synaptic devices
Tungsten oxides (WOx) are another great candidate
for memristive devices as a bio-inspired artificial
synapse. The advantage of WOx-based synaptic
devices includes accredited endurance (105 cycles),
CMOS compatibility and memorization and learning
functions. Here, we highlight the implementation of
this material in neuromorphic computing as a bio-
logical artificial synapse.
Synaptic behaviors and modeling of a WOx mem-
ristive device were reported by Chang [174] .The Pd/
WOx/W-structured memristor device shows reliable
synaptic operations with good endurance. The
synaptic devices can endure at least 105 potentiation/
depression pulses without degradation. Furthermore,
the memristor behavior was explained by a novel
model that takes both drift and diffusion effects into
consideration. Figure 14 presents the retention loss
Figure 12 a Micrograph of the multi-purpose neuromorphic chip
die showing the neuron tile array and the bias generator.
b Illustration of the hardware setup used to obtain measurements
in this paper. The presynaptic terminal of one neuron tile is
connected to the postsynaptic terminal of another neuron tile
through an off-chip TiO2-x memristor. A PC controls the digital
settings of the on-chip bias generator. Reprinted with permission
from Ref. [169].
Figure 13 Learning and forgetting of association in Pavlov’s dog
experiment realized with the three-neuron neural network.
Reprinted with permission from Ref [173].
J Mater Sci (2018) 53:8720–8746 8733
Author's personal copy
curve and forgetting memory in a human memory
curve from a paper by Chang [47]. It was found that
the trend of retention curve of a Pd/WOx/W-based
memristor and human forgetting curves greatly cor-
respond to each other and both can be fitted using a
stretched-exponential function (SEF). In addition, the
temperature and humidity impact on the perfor-
mance of a WOx memristive device was studied by
Meng et al. and Yong et al., respectively [175, 176].
According to the simulation results, the synaptic
weight of WOx memristive device decays faster as the
temperature increases due to higher oxygen vacan-
cies diffusion. In addition to the temperature, the
memristive effects of tungsten oxide are also highly
humidity dependent. The adsorbed moisture on the
surface of WO3 has resulted in decreasing conduc-
tances as the H cation induces an increase in barrier
heights.
Among other transition metal oxides, WOx is a
great candidate material for synaptic devices appli-
cation. For further exploring its application in
neuromorphic computing, enhancement of synaptic
operation time (endurance) is of importance.
AlOx-based synaptic devices
The resistive switching behavior in AlOx was first
reported by Hickmott [1]. AlOx is of interest in
RRAM materials due to its large band gap (* 9 eV)
and low RESET current (* lA). For neuromorphic
application, AlOx can also be used alone or stacked
with other RRAM materials to improve the unifor-
mity of the synaptic device characteristics.
AlOx-based resistive switching device was inves-
tigated by Wu [49] to serve as a potential electronic
synapse device. The 0.48 lm 9 0.48 lm Ti/AlOx/
TiN memory stack exhibited multi-level resistance
states when varying the compliance current levels or
the applied voltage amplitudes during pulse cycling.
In his work, around 1% resistance change per pulse
cycling was obtained and an STDP realization
scheme was proposed to implement the synaptic
device in large-scale hardware neuromorphic
Figure 14 a DC I–V curves
of a memristor studied.
b Schematic illustration of
oxygen vacancy diffusion.
c Retention loss curve of Pd/
WOx/W-based memristor.
d Forgetting memory of
human memory curve.
e Schematic illustration of
synaptic plasticity modulation.
Reprinted with permission
from Ref [47].
8734 J Mater Sci (2018) 53:8720–8746
Author's personal copy
computing system. In 2015, Prezioso experimentally
characterized Al2O3/TiO2-based memristors to
modulate the impact of conductance-dependent
conductance change on short-term dependent plas-
ticity [177]. The utilization of TiOx/Al2O3-based
memristor devices to emulate biological synaptic
behavior for building brain-inspired computers was
explored by Banerjee, W.et al. in 2017 [178]. The
proposed TiO2/Al2O3 synaptic device has the capa-
bility to switch from short-term memory (STM) to
long-term memory (LTM) by introducing a mezza-
nine state at medium memory (MM) as shown in
Fig. 15. In the paper, the dependence of retention
(decay time) on conductance levels (memory level)
was investigated and an experimentally proven
psychological model is presented.
Till now, the potential validity of Al2O3 materials
for electronic synapses had been experimentally
demonstrated. Nevertheless, the reports of sophisti-
cated Al2O3-based memristor as an artificial synapse
are limited. Therefore, further verification with dif-
ferent structures is required.
Other oxide-based synaptic devices
Besides the materials discussed above, a variety of
other materials has been studied to implement neural
network as a synaptic device. FeOx-based RRAM
with mixed-analogue–digital behavior was investi-
gated by Deng et al. [12] for a recurrent neural net-
work using the recursive least-squares algorithm.
Wang, C.H et al. reported the synaptic behavior of a
FeOx-based RRAM in which the dependence of
compliance current on conductance stability was
studied [179]. The application of CeO2-based RRAM
for highly energy efficient neuromorphic systems
was demonstrated by Kim et al. The Pt/CeO2/Pt
device exhibited the polarity-dependent and asym-
metric diode-type resistive switching [180], and the
detailed switching characteristics of artificial synap-
ses (potentiation and depression) were discussed.
A GdOx and Cu-doped MoOx stack with a platinum
top and bottom electrodes were reported by Choi
[45]. The weighted sum operation was carried out on
electrically modifiable synapse array circuit based on
the proposed stacks [181]. The biological synaptic
behavior was demonstrated by Chang through inte-
grating SiOx-based RRAM with Si diodes. The pro-
posed one-diode-one-resistor (1D-1R) architecture
not only avoids sneak-path issues and lowers
standby power consumption, but also helps to realize
STDP behaviors [182]. VOx is a well-known Mott
material which experience sharp and first-order
metal-to-insulator transition (MIT) at the around
68 �C [183]. The application of VOx as RRAM mate-
rials had been explored by Drisoll et al. [184] through
sol–gel technique. Nevertheless, most researches on
VOx so far focus on its use for select device which can
be integrated with RRAM device to mitigate sneak-
path current. The Pt/VO2/Pt selector has been inte-
grated with NiO unipolar RRAM by Lee et al. [185] in
2007 and ZrOx/HfOx bipolar RRAM by Son et al.
[186] in 2011. In 2016, 1S1R configuration of W/VO2/
Pt selection device and Ti/HfO2/Pt RRAM was
demonstrated by Kailiang et al. [187]. However,
thermal instability is a major challenge for VO2 for
practical applications [14].
Summary and outlook
In conclusion, we have outlined an overview of
oxide-based RRAM materials for the applications in
neuromorphic computing. Table 5 summarizes the
RRAM materials and their parameters discussed in
this review. Our work suggested that the neuromor-
phic approach with oxide-based RRAM devices is
promising. So far, PCMO, HfOx, TaOx, TiOx, NiOx,
WOx, AlOx, etc., materials have been demonstrated
for their application in oxide-based synaptic devices.
However, challenges still remain in specific materials,
Figure 15 Memory effects in the TiOx/Al2O3-based electronic
synaptic junction. The figure shows the conductance failure with
time. A linear scale plot is shown in the inset of the figure.
Reprinted with permission from Ref [178].
J Mater Sci (2018) 53:8720–8746 8735
Author's personal copy
e.g., poor symmetric weight update for PCMOx-
based synaptic devices and low endurance for WOx-
based synaptic devices. In general, there are two
main key challenges for overall oxide-based RRAM
materials. Although the inherent fault tolerance of
neural network models is able to mitigate the impact
of device variation to some extent, the improvement
of spatial variation and temporal variation turns out
to be one of the greatest challenges on a long-term
basis. In addition, improvement of reliability char-
acteristics of the RRAM synaptic devices is another
key challenge which is not well studied.
In future material research of neuromorphic com-
puting applications, the study of novel tunable
materials with enhanced properties for neuromor-
phic devices is one of the most exciting components.
One such example is MgO, which has a large band
gap that ensures sufficient band offsets, a high
dielectric constant that might potentially reduce
leakage current (thus improving device scalability), a
high thermal conductivity which minimizes self-
heating during operations and a large breakdown
field that might prevent the device from irreversible
damages. In addition, MgO-based magnetic tunnel
junction memristors had been proposed for imple-
mentation of synapses [188]. Further research is
required to expand the database of materials for
synaptic devices and neuromorphic applications.
Finally, to implement oxide-based RRAM devices
as a synapse in neuromorphic systems, it is impera-
tive to have an in-depth understanding of the meta-
plasticity mechanism and internal states of these
memristive devices. The underlying mechanisms
governing RRAM devices will inevitably be discov-
ered from investigations of conduction and resistive
switching mechanism via results from either experi-
ment or simulation. Further research work incorpo-
rating interactions between materials, device levels,
circuit designs and computing processes will
Table 5 Summary of RRAM materials in this review
Material On/
off
ratio
Endurance Energy
consumption
Multi-
levels
Switching
time
Neuromorphic computing applications
PCMO
[97, 136–146]
103 105 6 pJ 32 8 ns Pattern recognition
Human thought patterns
Neural fear-conditioning signal recognition
HfOx
[147–151]
102 106 10 pJ 8 10 ns Speech recognition
Visual pattern recognition
TaOx
[158–164]
10X 1012 10 fJ 200 105 ps Grayscale face classification
TiOx
[46, 166–170, 189]
105 2 X 106 sub-pJ 64 5 ns Pattern recognition in simulation
NiOx
[171–173, 185, 190, 191]
106 106 – 5 20 ns Learning and forgetting of association in Pavlov’s
dog experiment realized.
WOx
[47, 174–176, 192]
103 105 – 8 50 ns Human forgetting curves
AlOx
[49, 193–195]
106 105 1.5 pJ 10 10 ns Spike-timing-dependent plasticity realization
scheme.
FeOx
[179, 196, 197]
102 6 9 104 – 6 10 ns Long-term potentiation and long-term depression
demonstration
CeO2
[180, 198]
105 104 – 8 200 ns Synaptic potentiation and depression
characteristics demonstration
MoOx
[39, 199]
109 106 – 8 1 ms Weighted sum operation carried out on MoOx-
based electrically modifiable synapse array
circuit
SiOx
[182, 200–202]
107 106 – 4 100 ps Long-term potentiation and long-term depression
demonstration
8736 J Mater Sci (2018) 53:8720–8746
Author's personal copy
certainly speed up the realization of oxide-based
RRAM synapses for neuromorphic systems.
Acknowledgements
This work was supported by a RIE2020 AME-Pro-
grammatic Grant (Neuromorphic computing, No.
A1687b0033) and an Industry-IHL Partnership Pro-
gram (NRF2015-IIP001-001). WSL is a member of the
Singapore Spintronics Consortium (SG-SPIN).
References
[1] Hickmott TW (1962) Low-frequency negative resistance in
thin anodic oxide films. J Appl Phys 33(9):2669–2682.
https://doi.org/10.1063/1.1702530
[2] Liu SQ, Wu NJ, Ignatiev A (2000) Electric-pulse-induced
reversible resistance change effect in magnetoresistive
films. Appl Phys Lett 76(19):2749–2751. https://doi.org/10.
1063/1.126464
[3] Mandal S, El-Amin A, Alexander K, Rajendran B, Jha R
(2014) Novel synaptic memory device for neuromorphic
computing. Sci Rep. https://doi.org/10.1038/srep05333
[4] Lee MJ, Park Y, Kang BS, Ahn SE, Lee C, Kim K, Xianyu
WX, Stefanovich G, Lee JH, Chung SJ, Kim YH, Lee CS,
Park JB, Baek IG, Yoo IK (2007) 2-Stack 1D-1R cross-
point structure with oxide diodes as switch elements for
high density resistance RAM applications. In: International
electron devices meeting, 2007, pp 771–774. https://doi.
org/10.1109/iedm.2007.4419061
[5] Strukov DB, Snider GS, Stewart DR, Williams RS (2008)
The missing memristor found. Nature 453(7191):80–83.
https://doi.org/10.1038/nature06932
[6] Chevallier CJ, Siau CH, Lim SF, Namala SR, Matsuoka M,
Bateman BL, Rinerson D (2010) A 0.13 lm 64 Mb multi-
layered conductive metal-oxide memory. In: 2010 IEEE
international solid-state circuits conference—(ISSCC),
pp 260–261
[7] Liu TY, Yan TH, Scheuerlein R, Chen YC, Lee JK,
Balakrishnan G, Yee G, Zhang H, Yap A, Ouyang JW,
Sasaki T, Al-Shamma A, Chen CY, Gupta M, Hilton G,
Kathuria A, Lai V, Matsumoto M, Nigam A, Pai A, Pakhale
J, Siau CH, Wu XX, Yin YB, Nagel N, Tanaka Y, Higa-
shitani M, Minvielle T, Gorla C, Tsukamoto T, Yamaguchi
T, Okajima M, Okamura T, Takase S, Inoue H, Fasoli L
(2014) A 130.7-mm(2) 2-layer 32-Gb ReRAM memory
device in 24-nm technology. IEEE J Solid-State Circuits
49(1):140–153
[8] Fackenthal R, Kitagawa M, Otsuka W, Prall K, Mills D,
Tsutsui K, Javanifard J, Tedrow K, Tsushima T, Shibahara
Y, Hush G (2014) A 16 Gb ReRAM with 200 MB/s write
and 1 GB/s read in 27 nm technology. In: ISSCC digest
technical papers IEEE international, vol 57, pp 338–339
[9] Luo Q, Xu X, Liu H, Lv H, Gong T, Long S, Liu Q, Sun H,
Banerjee W, Li L, Gao J, Lu N, Liu M (2016) Super non-
linear RRAM with ultra-low power for 3D vertical nano-
crossbar arrays. Nanoscale 8(34):15629–15636. https://doi.
org/10.1039/C6NR02029A
[10] Mertens R (2017) Digitimes. http://www.digitimes.com/
newregister/join.asp?view=Article&DATEPUBLISH=2017/
06/05&PAGES=PB&SEQ=200. June 08, 2017
[11] Chang T-C, Chang K-C, Tsai T-M, Chu T-J, Sze SM (2016)
Resistance random access memory. Mater Today
19(5):254–264. https://doi.org/10.1016/j.mattod.2015.11.
009
[12] Deng L, Li G, Deng N, Wang D, Zhang Z, He W, Li H, Pei
J, Shi L (2015) Complex learning in bio-plausible mem-
ristive networks. Sci Rep 5:10684. https://doi.org/10.1038/
srep10684
[13] Advances and Trends of RRAM technology SemiconTai-
wan (2015)
[14] Wong HSP, Lee HY, Yu SM, Chen YS, Wu Y, Chen PS,
Lee B, Chen FT, Tsai MJ (2012) Metal-oxide RRAM. Proc
IEEE 100(6):1951–1970. https://doi.org/10.1109/Jproc.
2012.2190369
[15] Kuzum D, Yu SM, Wong HSP (2013) Synaptic electronics:
materials, devices and applications. Nanotechnology.
https://doi.org/10.1088/0957-4484/24/38/382001
[16] Akinaga H, Shima H (2010) Resistive random access
memory (ReRAM) based on metal oxides. Proc IEEE
98(12):2237–2251. https://doi.org/10.1109/Jproc.2010.
2070830
[17] Waser R, Aono M (2007) Nanoionics-based resistive
switching memories. Nat Mater 6(11):833–840. https://doi.
org/10.1038/nmat2023
[18] Chi P, Li SC, Xu C, Zhang T, Zhao JS, Liu YP, Wang Y,
Xie Y (2016) PRIME: a novel processing-in-memory
architecture for neural network computation in ReRAM-
based main memory. In: Conference proceedings of inter-
national symposium coastal engineering, pp 27–39. https://
doi.org/10.1109/isca.2016.13
[19] Qian K, Nguyen VC, Chen TP, Lee PS (2016) Novel
concepts in functional resistive switching memories.
J Mater Chem C 4(41):9637–9645. https://doi.org/10.1039/
c6tc03447k
[20] Tsunoda K, Kinoshita K, Noshiro H, Yarnazaki Y, Iizuka T,
Ito Y, Takahashi A, Okano A, Sato Y, Fukano T, Aoki M,
Sugiyama Y (2007) Low power and high speed switching
J Mater Sci (2018) 53:8720–8746 8737
Author's personal copy
of Ti-doped NiOReRAM under the unipolar voltage source
of less than 3 V. In: International electron devices meeting,
pp 767–770. https://doi.org/10.1109/iedm.2007.4419060
[21] Wang ZR, Zhu WG, Du AY, Wu L, Fang Z, Tran XA, Liu
WJ, Zhang KL, Yu HY (2012) Highly uniform, self-com-
pliance, and forming-free ALD HfO2-based RRAM with
Ge doping. IEEE Trans Electron Dev 59(4):1203–1208.
https://doi.org/10.1109/Ted.2012.2182770
[22] Magyari-Kope B, Duncan D, Zhao L, Nishi Y (2016)
Doping technology for RRAM—opportunities and chal-
lenges. In: International symposium on VLSI technology
[23] Yu S, Wu Y, Chai Y, Provine J, Wong HSP (2011) Char-
acterization of switching parameters and multilevel capa-
bility in HfOx/AlOx bi-layer RRAM devices. In: VLSI
technology (VLSIT), 2011 symposium on, pp 106–107
[24] Park SG, Yang MK, Ju H, Seong DJ, Lee JM, Kim E, Jung
S, Zhang L, Shin YC, Baek IG, Choi J, Kang HK, Chung C
(2012) A non-linear ReRAM cell with sub-1 mu a ultralow
operating current for high density vertical resistive memory
(VRRAM). In: 2012 IEEE international electron devices
meeting (IEDM)
[25] Yun JB, Kim S, Seo S, Lee MJ, Kim DC, Ahn SE, Park Y,
Kim J, Shin H (2007) Random and localized resistive
switching observation in Pt/NiO/Pt. Phys Status Solidi-R
1(6):280–282. https://doi.org/10.1002/pssr.200701205
[26] Ho PWC, Hatem FO, Almurib HAF, Kumar TN (2016)
Comparison between Pt/TiO2/Pt and Pt/TaOX/TaOY/Pt
based bipolar resistive switching devices. J Semicond.
https://doi.org/10.1088/1674-4926/37/6/064001
[27] Lei XY, Liu HX, Gao HX, Yang HN, Wang GM, Long SB,
Ma XH, Liu M (2014) Resistive switching characteristics of
Ti/ZrO2/Pt RRAM device. Chin Phys B. https://doi.org/10.
1088/1674-1056/23/11/117305
[28] Long SB, Lian XJ, Ye TC, Cagli C, Perniola L, Miranda E,
Liu M, Sune J (2013) Cycle-to-cycle intrinsic RESET
statistics in HfO2-based unipolar RRAM devices. IEEE
Electron Dev Lett 34(5):623–625. https://doi.org/10.1109/
Led.2013.2251314
[29] Yoshida C, Tsunoda K, Noshiro H, Sugiyama Y (2007)
High speed resistive switching in Pt/TiO2/TiN film for
nonvolatile memory application. Appl Phys Lett. https://
doi.org/10.1063/1.2818691
[30] Xu N, Liu LF, Sun X, Chen C, Wang Y, Han DD, Liu XY,
Han RQ, Kang JF, Yu B (2008) Bipolar switching behavior
in TiN/ZnO/Pt resistive nonvolatile memory with fast
switching and long retention. Semicond Sci Technol.
https://doi.org/10.1088/0268-1242/23/7/075019
[31] Kim S, Lee D, Park J, Jung S, Lee W, Shin J, Woo J, Choi
G, Hwang H (2012) Defect engineering: reduction effect of
hydrogen atom impurities in HfO2-based resistive-
switching memory devices. Nanotechnology. https://doi.
org/10.1088/0957-4484/23/32/325702
[32] Ma G, Tang X, Zhang H, Zhong Z, Li X, Li J, Su H (2017)
Ultra-high ON/OFF ratio and multi-storage on NiO resis-
tive switching device. J Mater Sci 52(1):238–246. https://
doi.org/10.1007/s10853-016-0326-5
[33] Sarkar PK, Prajapat M, Barman A, Bhattacharjee S, Roy A
(2016) Multilevel resistance state of Cu/La2O3/Pt forming-
free switching devices. J Mater Sci 51(9):4411–4418.
https://doi.org/10.1007/s10853-016-9753-6
[34] Tsai CL, Xiong F, Pop E, Shim M (2013) Resistive random
access memory enabled by carbon nanotube crossbar
electrodes. ACS Nano 7(6):5360–5366. https://doi.org/10.
1021/nn401212p
[35] Wu Y, Chai Y, Chen H-Y, Yu S, Wong H-SP (2011)
Resistive switching AlOx-based memory with CNT elec-
trode for ultra-low switching current and high density
memory application. In: VLSI technology (VLSIT), 2011
Symposium on
[36] Munoz-Gorriz J, Acero MC, Gonzalez MB, Campabadal F
(2017) Top electrode dependence of the resistive switching
behavior in HfO2/n(?)Si-based devices. In: Spanish con-
ference electron
[37] Sohn J, Lee S, Jiang ZZ, Chen HY, Wong HSP (2014)
Atomically thin graphene plane electrode for 3D RRAM.
In: 2014 IEEE international electron devices meeting
(IEDM)
[38] Cheng CH, Chin A, Yeh FS (2011) Ultralow switching
energy Ni/GeOx/HfON/TaN RRAM. IEEE Electron Device
Lett 32(3):366–368. https://doi.org/10.1109/Led.2010.
2095820
[39] Yoon J, Choi H, Lee D, Park JB, Lee J, Seong DJ, Ju Y,
Chang M, Jung S, Hwang H (2009) Excellent switching
uniformity of Cu-Doped MoOx/GdOx bilayer for non-
volatile memory applications. IEEE Electron Device Lett
30(5):457–459. https://doi.org/10.1109/Led.2009.2015687
[40] Yu SM, Gao B, Fang Z, Yu HY, Kang JF, Wong HSP
(2013) A low energy oxide-based electronic synaptic
device for neuromorphic visual systems with tolerance to
device variation. Adv Mater 25(12):1774–1779. https://doi.
org/10.1002/adma.201203680
[41] Lee AR, Bae YC, Baek GH, Im HS, Hong JP (2013) Multi-
level resistive switching observations in asymmetric Pt/
Ta2O5 - x/TiOxNy/TiN/Ta2O5 - x/Pt multilayer config-
urations. Appl Phys Lett. https://doi.org/10.1063/1.
4818129
[42] Kumar D, Aluguri R, Chand U, Tseng TY (2017) Metal
oxide resistive switching memory: materials, properties and
switching mechanisms. Ceram Int 43:S547–S556. https://
doi.org/10.1016/j.ceramint.2017.05.289
8738 J Mater Sci (2018) 53:8720–8746
Author's personal copy
[43] Drachman DA (2005) Do we have brain to spare? Neu-
rology 64(12):2004–2005. https://doi.org/10.1212/01.Wnl.
0000166914.38327.Bb
[44] Suri M, Bichler O, Querlioz D, Palma G, Vianello E,
Vuillaume D, Gamrat C, DeSalvo B (2012) CBRAM
devices as binary synapses for low-power stochastic neu-
romorphic systems: auditory (Cochlea) and visual (Retina)
cognitive processing applications. In: 2012 IEEE interna-
tional electron devices meeting (IEDM)
[45] Choi H, Jung H, Lee J, Yoon J, Park J, Seong DJ, Lee W,
Hasan M, Jung GY, Hwang H (2009) An electrically
modifiable synapse array of resistive switching memory.
Nanotechnology. https://doi.org/10.1088/0957-4484/20/34/
345201
[46] Seo K, Kim I, Jung S, Jo M, Park S, Park J, Shin J, Biju KP,
Kong J, Lee K, Lee B, Hwang H (2011) Analog memory
and spike-timing-dependent plasticity characteristics of a
nanoscale titanium oxide bilayer resistive switching device.
Nanotechnology. https://doi.org/10.1088/0957-4484/22/25/
254023
[47] Chang T, Jo SH, Lu W (2011) Short-term memory to long-
term memory transition in a nanoscale memristor. ACS
Nano 5(9):7669–7676. https://doi.org/10.1021/nn202983n
[48] Yang R, Terabe K, Liu GQ, Tsuruoka T, Hasegawa T,
Gimzewski JK, Aono M (2012) On-demand nanodevice
with electrical and neuromorphic multifunction realized by
local ion migration. ACS Nano 6(11):9515–9521. https://
doi.org/10.1021/nn302510e
[49] Wu Y, Yu SM, Wong HSP, Chen YS, Lee HY, Wang SM,
Gu PY, Chen F, Tsai MJ (2012) AlOx-based resistive
switching device with gradual resistance modulation for
neuromorphic device application. In: IEEE international
memory workshop
[50] Suri M, Bichler O, Hubert Q, Perniola L, Sousa V, Jahan C,
Vuillaume D, Gamrat C, DeSalvo B (2012) Interface
engineering of PCM for improved synaptic performance in
neuromorphic systems. In: IEEE international memory
workshop
[51] Eryilmaz SB, Kuzum D, Jeyasingh R, Kim S, BrightSky M,
Lam C, Wong HSP (2014) Brain-like associative learning
using a nanoscale non-volatile phase change synaptic
device array. Front Neurosci-Switz. https://doi.org/10.3389/
fnins.2014.00205
[52] Garbin D, Suri M, Bichler O, Querlioz D, Gamrat C,
DeSalvo B (2013) Probabilistic neuromorphic system using
binary phase-change memory (PCM) synapses: detailed
power consumption analysis. In: 2013 13th IEEE confer-
ence on nanotechnology (IEEE-Nano), pp 91–94
[53] Jo SH, Chang T, Ebong I, Bhadviya BB, Mazumder P, Lu
W (2010) Nanoscale memristor device as synapse in
neuromorphic systems. Nano Lett 10(4):1297–1301.
https://doi.org/10.1021/nl904092h
[54] Kang DH, Jun HG, Ryoo KC, Jeong H, Sohn H (2015)
Emulation of spike-timing dependent plasticity in nano-
scale phase change memory. Neurocomputing
155:153–158. https://doi.org/10.1016/j.neucom.2014.12.
036
[55] Kim S, Ishii M, Lewis S, Perri T, BrightSky M, Kim W,
Jordan R, Burr GW, Sosa N, Ray A, Han JP, Miller C,
Hosokawa K, Lam C (2015) NVM neuromorphic core with
64 k-cell (256-by-256) phase change memory synaptic
array with on-chip neuron circuits for continuous in situ
learning. In: 2015 IEEE international electron devices
meeting (IEDM)
[56] Shelby RM, Burr GW, Boybat I, Di Nolfo C (2015) Non-
volatile memory as hardware synapse in neuromorphic
computing: a first look at reliability issues. In: 2015 IEEE
international reliability physics symposium (IRPS)
[57] Suri M, Bichler O, Querlioz D, Cueto O, Perniola L, Sousa
V, Vuillaume D, Gamrat C, DeSalvo B (2011) Phase
change memory as synapse for ultra-dense neuromorphic
systems: application to complex visual pattern extraction.
In: 2011 IEEE international electron devices meeting
(IEDM)
[58] Suri M, Bichler O, Querlioz D, Traore B, Cueto O, Perniola
L, Sousa V, Vuillaume D, Gamrat C, DeSalvo B (2012)
Physical aspects of low power synapses based on phase
change memory devices. J Appl Phys. https://doi.org/10.
1063/1.4749411
[59] Zhong YP, Li Y, Xu L, Miao XS (2015) Simple square
pulses for implementing spike-timing-dependent plasticity
in phase-change memory. Phys Status Solidi-R
9(7):414–419. https://doi.org/10.1002/pssr.201510150
[60] Kuzum D, Jeyasingh RGD, Lee B, Wong HSP (2012)
Nanoelectronic programmable synapses based on phase
change materials for brain-inspired computing. Nano Lett
12(5):2179–2186. https://doi.org/10.1021/nl201040y
[61] Kuzum D, Jeyasingh RGD, Wong HSP (2011) Energy
efficient programming of nanoelectronic synaptic devices
for large-scale implementation of associative and temporal
sequence learning. In: 2011 IEEE international electron
devices meeting (IEDM)
[62] Sharad M, Augustine C, Roy K (2012) Boolean and non-
boolean computation with spin devices. In: 2012 IEEE
international electron devices meeting (IEDM)
[63] Sangwan VK, Jariwala D, Kim IS, Chen KS, Marks TJ,
Lauhon LJ, Hersam MC (2015) Gate-tunable memristive
phenomena mediated by grain boundaries in single-layer
MoS2. Nat Nanotechnol 10(5):403–406. https://doi.org/10.
1038/Nnano.2015.56
J Mater Sci (2018) 53:8720–8746 8739
Author's personal copy
[64] Sengupta A, Al Azim Z, Fong XY, Roy K (2015) Spin-orbit
torque induced spike-timing dependent plasticity. Appl
Phys Lett. https://doi.org/10.1063/1.4914111
[65] Sengupta A, Parsa M, Han B, Roy K (2016) Probabilistic
deep spiking neural systems enabled by magnetic tunnel
junction. IEEE Trans Electron Dev 63(7):2963–2970.
https://doi.org/10.1109/Ted.2016.2568762
[66] Sengupta A, Roy K (2016) Short-term plasticity and long-
term potentiation in magnetic tunnel junctions: towards
volatile synapses. Phys Rev Appl. https://doi.org/10.1103/
PhysRevApplied.5.024012
[67] Adachi M, Seki M, Yamahara H, Nasu H, Tabata H (2015)
Long-term potentiation of magnonic synapses by photo-
controlled spin current mimicked in reentrant spin-glass
garnet ferrite Lu3Fe5-2xCoxSixO12 thin films. Appl Phys
Express. https://doi.org/10.7567/Apex.8.043002
[68] Roska T, Horvath A, Stubendek A, Corinto F, Csaba G,
Porod W, Shibata T, Bourianoff G (2012) An associative
memory with oscillatory CNN arrays using spin torque
oscillator cells and spin-wave interactions architecture and
end-to-end simulator. In: International work cell nano
[69] Fujita O, Amemiya Y (1993) A floating-gate analog
memory device for neural networks. IEEE Trans Electron
Dev 40(11):2029–2055. https://doi.org/10.1109/16.239745
[70] Hafliger P, Rasche C (1999) Floating gate analog memory
for parameter and variable storage in a learning silicon
neuron. In: ISCAS ‘99: proceedings of the 1999 IEEE
international symposium on circuits and systems, vol 2,
pp 416–419
[71] Harrison RR, Bragg JA, Hasler P, Minch BA, Deweerth SP
(2001) A CMOS programmable analog memory-cell array
using floating-gate circuits. IEEE Trans Circuits-II
48(1):4–11. https://doi.org/10.1109/82.913181
[72] Brink S, Nease S, Hasler P (2013) Computing with net-
works of spiking neurons on a biophysically motivated
floating-gate based neuromorphic integrated circuit. Neural
Netw 45:39–49. https://doi.org/10.1016/j.neunet.2013.02.
011
[73] Diorio C, Hasler P, Minch BA, Mead CA (1997) A floating-
gate MOS learning array with locally computed weight
updates. IEEE Trans Electron Dev 44(12):2281–2289.
https://doi.org/10.1109/16.644652
[74] Diorio C, Hasler P, Minch BA, Mead C (1997) A com-
plementary pair of four-terminal silicon synapses. Analog
Integr Circ Signal Process 13(1–2):153–166. https://doi.
org/10.1023/A:1008244314595
[75] Rahimi K, Diorio C, Hernandez C, Brockhausen MD
(2002) A simulation model for floating-gate MOS synapse
transistors. In: 2002 IEEE international symposium on
circuits and systems, vol II, proceedings, pp 532–535
[76] Diorio C, Hasler P, Minch A, Mead CA (1996) A single-
transistor silicon synapse. IEEE Trans Electron Dev
43(11):1972–1980. https://doi.org/10.1109/16.543035
[77] Cauwenberghs G, Neugebauer CF, Agranat AJ, Yariv A
(1990) Large scale optoelectronic integration of asyn-
chronous analog neural networks. In: International neural
network conference: July 9–13, 1990 Palais Des Congres—
Paris—France. Springer, Netherlands, pp 551–554. https://
doi.org/10.1007/978-94-009-0643-3_1
[78] Burla M, Wang X, Li M, Chrostowski L, Azana J (2016)
Wideband dynamic microwave frequency identification
system using a low-power ultracompact silicon photonic
chip. Nat Commun. https://doi.org/10.1038/ncomms13004
[79] Rietman EA, Frye RC, Wong CC, Kornfeld CD (1989)
Amorphous-silicon photoconductive arrays for artificial
neural networks. Appl Opt 28(16):3474–3478
[80] Livingston DL, Albin S, Park SM (1991) A new method for
storing weights in analog neural hardware. In: IEEE pro-
ceedings of the Southeastcon 91, vols 1 and 2, pp 86–88.
https://doi.org/10.1109/secon.1991.147710
[81] Maier P, Hartmann F, Emmerling M, Schneider C, Kamp
M, Hofling S, Worschech L (2016) Electro-photo-sensitive
memristor for neuromorphic and arithmetic computing.
Phys Rev Appl. https://doi.org/10.1103/PhysRevApplied.5.
054011
[82] Boukhobza J, Rubini S (2012) Flashing in the memory
hierarchy: an overview on flash memory internals
[83] Kotecki DE (1997) A review of high dielectric materials for
DRAM capacitors. Integr Ferroelectr 16(1–4):1–19. https://
doi.org/10.1080/10584589708013025
[84] Pavan P, Bez R, Olivo P, Zanoni E (1997) Flash memory
cells: an overview. Proc IEEE 85(8):1248–1271. https://doi.
org/10.1109/5.622505
[85] Sethi P, Krishnia S, Gan WL, Kholid FN, Tan FN, Maddu
R, Lew WS (2017) Bi-directional high speed domain wall
motion in perpendicular magnetic anisotropy Co/Pt double
stack structures. Sci Rep. https://doi.org/10.1038/s41598-
017-05409-7
[86] Zhang SF, Gan WL, Kwon J, Luo FL, Lim GJ, Wang JB,
Lew WS (2016) Highly efficient domain walls injection in
perpendicular magnetic anisotropy nanowire. Sci Rep.
https://doi.org/10.1038/srep24804
[87] Apalkov D, Dieny B, Slaughter JM (2016) Magnetoresis-
tive random access memory. Proc IEEE
104(10):1796–1830. https://doi.org/10.1109/Jproc.2016.
2590142
[88] Hsu J (2014) IBM’s new brain. IEEE Spectr 51(10):17–19
[89] Yu SM, Chen HY, Gao B, Kang JF, Wong HSP (2013)
HfOx-based vertical resistive switching random access
memory suitable for bit-cost-effective three-dimensional
8740 J Mater Sci (2018) 53:8720–8746
Author's personal copy
cross-point architecture. ACS Nano 7(3):2320–2325.
https://doi.org/10.1021/nn305510u
[90] Zhang ZP, Wu Y, Wong HSP, Wong SS (2013) Nanometer-
scale HfOx RRAM. IEEE Electron Dev Lett
34(8):1005–1007. https://doi.org/10.1109/Led.2013.
2265404
[91] Yu SM, Gao B, Fang Z, Yu HY, Kang JF, Wong HSP
(2013) Stochastic learning in oxide binary synaptic device
for neuromorphic computing. Front Neurosci-Switz. https://
doi.org/10.3389/fnins.2013.00186
[92] Gao B, Bi YJ, Chen HY, Liu R, Huang P, Chen B, Liu LF,
Liu XY, Yu SM, Wong HSP, Kang JF (2014) Ultra-low-
energy three-dimensional oxide-based electronic synapses
for implementation of robust high-accuracy neuromorphic
computation systems. ACS Nano 8(7):6998–7004. https://
doi.org/10.1021/nn501824r
[93] Liu JC, Hsu CW, Wang IT, Hou TH (2015) Categorization
of multilevel-cell storage-class memory: an RRAM exam-
ple. IEEE Trans Electron Dev 62(8):2510–2516. https://doi.
org/10.1109/Ted.2015.2444663
[94] Hsu CW, Hou TH, Chen MC, Wang IT, Lo CL (2013)
Bipolar Ni/TiO2/HfO2/Ni RRAM with multilevel states
and self-rectifying characteristics. IEEE Electron Device
Lett 34(7):885–887. https://doi.org/10.1109/Led.2013.
2264823
[95] Hou TH (2015) 3D RRAM-based synaptic network with
low programming energy. In: 5th stanford-IMEC Interna-
tional RRAM Workshop, Leuven, Belgium, pp 24–25
[96] Burr GW, Shelby RM, Sebastian A, Kim S, Kim S, Sidler
S, Virwani K, Ishii M, Narayanan P, Fumarola A, Sanches
LL, Boybat I, Le Gallo M, Moon K, Woo J, Hwang H,
Leblebici Y (2017) Neuromorphic computing using non-
volatile memory. Adv Phys-X 2(1):89–124. https://doi.org/
10.1080/23746149.2016.1259585
[97] Lee D, Park J, Moon K, Jang J-W, Park S, Chu M, Kim J,
Noh J, Jeon M, Lee B, Lee B, Lee B-G, Hwang H (2015)
Oxide based nanoscale analog synapse device for neural
signal recognition system. In: Electron devices meeting
(IEDM), 2015 IEEE international. https://doi.org/10.1109/
iedm.2015.7409628
[98] Slutsky I, Abumaria N, Wu L-J, Huang C, Zhang L, Li B,
Zhao X, Govindarajan A, Zhao M-G, Zhuo M, Tonegawa
S, Liu G (2010) Enhancement of learning and memory by
elevating brain magnesium. Neuron 65(2):165–177. https://
doi.org/10.1016/j.neuron.2009.12.026
[99] Mayford M, Siegelbaum SA, Kandel ER (2012) Synapses
and memory storage. Cold Spring Harb Perspect Biol
4(6):a005751. https://doi.org/10.1101/cshperspect.a005751
[100] Okano H, Hirano T, Balaban E (2000) Learning and
memory. Proc Natl Acad Sci 97(23):12403–12404
[101] Shaw GL (1986) Donald Hebb: the organization of
behavior. In: Palm G, Aertsen A (eds) Brain theory: pro-
ceedings of the first trieste meeting on brain theory, October
1–4, 1984. Springer, Berlin, pp 231–233. https://doi.org/10.
1007/978-3-642-70911-1_15
[102] Bi GQ, Poo MM (1998) Synaptic modifications in cultured
hippocampal neurons: dependence on spike timing,
synaptic strength, and postsynaptic cell type. J Neurosci
18(24):10464–10472
[103] Zhang LI, Tao HW, Holt CE, Harris WA, Poo MM (1998)
A critical window for cooperation and competition among
developing retinotectal synapses. Nature 395(6697):37–44
[104] Markram H, Lubke J, Frotscher M, Sakmann B (1997)
Regulation of synaptic efficacy by coincidence of postsy-
naptic APs and EPSPs. Science 275(5297):213–215.
https://doi.org/10.1126/science.275.5297.213
[105] Ohno T, Hasegawa T, Tsuruoka T, Terabe K, Gimzewski
JK, Aono M (2011) Short-term plasticity and long-term
potentiation mimicked in single inorganic synapses. Nat
Mater 10(8):591–595. https://doi.org/10.1038/Nmat3054
[106] Chen XR, Feng J, Ma HL (2015) The resistive switching
characteristics in tantalum oxide-based RRAM device via
combining high-temperature sputtering with plasma oxi-
dation. In: 2015 15th non-volatile memory technology
symposium (NVMTS)
[107] Li YY, Liu ZT, Tan TT (2014) Resistive switching behavior
of hafnium oxide thin film grown by magnetron sputtering.
Rare Metal Mat Eng 43(1):24–27
[108] Wu Y, Lee B, Wong H-SP (2010) Ultra-low power Al2O3-
based RRAM with 1 lA reset current. In: VLSI technology
systems and applications (VLSI-TSA), 2010 international
symposium on
[109] Chang KC, Tsai TM, Chang TC, Syu YE, Chuang SL, Li
CH, Gan DS, Sze SM (2012) The effect of silicon oxide
based RRAM with tin doping. Electrochem Solid-State Lett
15(3):H65–H68. https://doi.org/10.1149/2.013203esl
[110] Wang SY, Tsai CH, Lee DY, Lin CY, Lin CC, Tseng TY
(2011) Improved resistive switching properties of Ti/ZrO2/
Pt memory devices for RRAM application. Microelectron
Eng 88(7):1628–1632. https://doi.org/10.1016/j.mee.2010.
11.058
[111] Liu KC, Tzeng WH, Chang KM, Wu CH (2010) The effect
of plasma deposition on the electrical characteristics of Pt/
HfOx/TiN RRAM device. Surf Coat Technol 205:S379–
S384. https://doi.org/10.1016/j.surfcoat.2010.08.043
[112] Simanjuntak FM, Panda D, Tsai T-L, Lin C-A, Wei K-H,
Tseng T-Y (2015) Enhancing the memory window of AZO/
ZnO/ITO transparent resistive switching devices by mod-
ulating the oxygen vacancy concentration of the top
J Mater Sci (2018) 53:8720–8746 8741
Author's personal copy
electrode. J Mater Sci 50(21):6961–6969. https://doi.org/
10.1007/s10853-015-9247-y
[113] Wang J, Nonnenmann SS (2017) Area-dependent electro-
forming and switching polarity reversal across TiO2/
Nb:SrTiO3 oxide interfaces. J Mater Sci
52(11):6469–6475. https://doi.org/10.1007/s10853-017-
0882-3
[114] Wu Y, Lee B, Wong HSP (2010) Al2O3-Based RRAM
using atomic layer deposition (ALD) with 1-mu a RESET
current. IEEE Electron Device Lett 31(12):1449–1451.
https://doi.org/10.1109/Led.2010.2074177
[115] Egorov KV, Lebedinskii YY, Markeev AM, Orlov OM
(2015) Full ALD Ta2O5-based stacks for resistive random
access memory grown with in vacuo XPS monitoring. Appl
Surf Sci 356:454–459. https://doi.org/10.1016/j.apsusc.
2015.07.217
[116] Sahu VK, Misra P, Ajimsha RS, Das AK, Joshi MP,
Kukreja LM (2016) Resistive memory switching in ultra-
thin TiO2 films grown by atomic layer deposition. In: AIP
conference proceedings vol 1731. https://doi.org/10.1063/1.
4948104
[117] Niu G, Kim HD, Roelofs R, Perez E, Schubert MA,
Zaumseil P, Costina I, Wenger C (2016) Material insights of
HfO2-based integrated 1-transistor-1-resistor resistive ran-
dom access memory devices processed by batch atomic
layer deposition. Sci Rep. https://doi.org/10.1038/
srep28155
[118] Misra P, Das AK, Kukreja LM (2010) Switching charac-
teristics of ZnO based transparent resistive random access
memory devices grown by pulsed laser deposition. Phys
Status Solidi C 7(6):1718–1720. https://doi.org/10.1002/
pssc.200983244
[119] Wu Z, Zhu J (2017) Enhanced unipolar resistive switching
characteristics of Hf0.5Zr0.5O2 thin films with high ON/
OFF ratio. Materials. https://doi.org/10.3390/ma10030322
[120] Parreira P, Paterson GW, McVitie S, MacLaren DA (2016)
Stability, bistability and instability of amorphous ZrO2
resistive memory devices. J Phys D Appl Phys. https://doi.
org/10.1088/0022-3727/49/9/095111
[121] Xu ZD, Yu LN, Wu Y, Dong C, Deng N, Xu XG, Miao J,
Jiang Y (2015) Low-energy resistive random access
memory devices with no need for a compliance current. Sci
Rep. https://doi.org/10.1038/srep10409
[122] Hsu ST, Zhuang WW, Li TK, Pan W, Ignatiev A, Papa-
gianni C, Wu NJ (2005) RRAM switching mechanism. In:
2005 Non-volatile memory technology symposium, pro-
ceedings, pp 121–124
[123] Preparation method of tungsten oxide variable-resistance
material in resistive random access memory (2014) Google
Patents
[124] Rodriguez R, Poulter B, Gonzalez M, Ali F, Lau LD,
Mangun M (2017) The effect of tin on PECVD-deposited
germanium sulfide thin films for resistive RAM devices.
Mater Sci Appl 08:188–196
[125] Yao J, Sun ZZ, Zhong L, Natelson D, Tour JM (2010)
Resistive switches and memories from silicon oxide. Nano
Lett 10(10):4105–4110. https://doi.org/10.1021/nl102255r
[126] Mustaqima M, Yoo P, Huang W, Lee BW, Liu CL (2015)
Regulation of the forming process and the set voltage dis-
tribution of unipolar resistance switching in spin-coated
CoFe2O4 thin films. Nanoscale Res Lett. https://doi.org/10.
1186/s11671-015-0876-5
[127] Xu HT, Wu CJ, Zhao XH, Jung RJ, Li Y, Liu CL (2017)
Improved resistance switching stability in Fe-doped ZnO
thin films through pulsed magnetic field annealing.
Nanoscale Res Lett. https://doi.org/10.1186/s11671-017-
1949-4
[128] Zhang F, Zhuang WW, Evans DR, Hsu ST (2004) High
temperature annealing of spin coated Pr1-xCaxMnO3 thim
film for RRAM application. Google patents
[129] Chang YC, Wei CY, Wang YH (2012) Sol-gel barium
titanate-based RRAM by inserting graphene oxide inter-
layer. 2012 IEEE international conference on electron
devices and solid state circuit (EDSSC)
[130] Wang Y, Chen B, Liu D, Gao B, Liu LF, Liu X, Kang J
(2014) Solution processed resistive random access memory
devices for transparent solid-state circuit systems. In: MRS
online proceedings library archive, vol 1633. https://doi.
org/10.1557/opl.2014.252
[131] Lee C, Kim I, Choi W, Shin H, Cho J (2009) Resistive
switching memory devices composed of binary transition
metal oxides using sol–gel chemistry. Langmuir
25(8):4274–4278
[132] Kim HD, Yun MJ, Lee JH, Kim KH, Kim TG (2014)
Transparent multi-level resistive switching phenomena
observed in ITO/RGO/ITO memory cells by the sol-gel
dip-coating method. Sci Rep. https://doi.org/10.1038/
srep04614
[133] Niu G, Calka P, Auf der Maur M, Santoni F, Guha S,
Fraschke M, Hamoumou P, Gautier B, Perez E, Walczyk C,
Wenger C, Di Carlo A, Alff L, Schroeder T (2016) Geo-
metric conductive filament confinement by nanotips for
resistive switching of HfO2-RRAM devices with high
performance. Sci Rep 6:25757. https://doi.org/10.1038/
srep25757
[134] Siddiqui G-U-D, Ali J, Doh Y-H, Choi KH (2016) Fabri-
cation of zinc stannate based all-printed resistive switching
device. Mater Lett 166(Supplement C):311–316. https://
doi.org/10.1016/j.matlet.2015.12.045
8742 J Mater Sci (2018) 53:8720–8746
Author's personal copy
[135] Asamitsu A, Tomioka Y, Kuwahara H, Tokura Y (1997)
Current switching of resistive states in magnetoresistive
manganites. Nature 388:50. https://doi.org/10.1038/40363
[136] Chen X, Wu NJ, Ignatiev A (2005) Perovskite RRAM
devices with metal/insulator/PCMO/metal heterostructures.
In: 2005 Non-volatile memory technology symposium,
proceedings, pp 125–128
[137] Liu KC, Tzeng WH, Chang KM, Chan YC, Kuo CC,
Cheng CW (2010) Transparent resistive random access
memory (T-RRAM) based on Gd2O3 film and its resistive
switching characteristics. In: INEC: 2010 3rd international
nanoelectronics conference, vols 1 and 2, pp 898–899
[138] Park S, Kim H, Choo M, Noh J, Sheri A, Jung S, Seo K,
Park J, Kim S, Lee W, Shin J, Lee D, Choi G, Woo J, Cha
E, Jang J, Park C, Jeon M, Lee B, Lee BH, Hwang H
(2012) RRAM-based synapse for neuromorphic system
with pattern recognition function. In: 2012 IEEE interna-
tional electron devices meeting (IEDM)
[139] Sangsu P, Jinwoo N, Myung-lae C, Ahmad Muqeem S,
Man C, Young-Bae K, Chang Jung K, Moongu J, Byung-
Geun L, Byoung Hun L, Hyunsang H (2013) Nanoscale
RRAM-based synaptic electronics: toward a neuromorphic
computing device. Nanotechnology 24(38):384009
[140] Park S, Sheri A, Kim J, Noh J, Jang J, Jeon M, Lee B, Lee
BR, Lee BH, Hwang H (2013) Neuromorphic speech sys-
tems using advanced ReRAM-based synapse. In: 2013
IEEE international electron devices meeting (IEDM)
[141] Sangsu P, Manzar S, Jinwoo N, Daeseuk L, Kibong M,
Jiyong W, Byoung HL, Hyunsang H (2014) A nitrogen-
treated memristive device for tunable electronic synapses.
Semicond Sci Technol 29(10):104006
[142] Moon K, Park S, Jang J, Lee D, Woo J, Cha E, Lee S, Park
J, Song J, Koo Y, Hwang H (2014) Hardware implemen-
tation of associative memory characteristics with analogue-
type resistive-switching device. Nanotechnology. https://
doi.org/10.1088/0957-4484/25/49/495204
[143] Lee D, Moon K, Park J, Park S, Hwang H (2015) Trade-off
between number of conductance states and variability of
conductance change in Pr0.7Ca0.3MnO3-based synapse
device. Appl Phys Lett. https://doi.org/10.1063/1.4915924
[144] Moon K, Cha E, Park J, Gi S, Chu M, Baek K, Lee B, Oh
SH, Hwang H (2016) Analog synapse device with 5-b
MLC and improved data retention for neuromorphic sys-
tem. IEEE Electron Device Lett 37(8):1067–1070. https://
doi.org/10.1109/Led.2016.2583545
[145] Moon K, Cha E, Lee D, Jang J, Park J, Hwang H (2016)
ReRAM-based analog synapse and IMT neuron device for
neuromorphic system. In: International symposium VLSI
technology
[146] Yu S (2017) Neuro-inspired computing using resistive
synaptic devices. Springer, Berlin
[147] Yu SM, Wu Y, Jeyasingh R, Kuzum DG, Wong HSP (2011)
An electronic synapse device based on metal oxide resistive
switching memory for neuromorphic computation. IEEE
Trans Electron Dev 58(8):2729–2737. https://doi.org/10.
1109/Ted.2011.2147791
[148] Gao B, Chen B, Zhang FF, Liu LF, Liu XY, Kang JF, Yu
HY, Yu B (2013) A novel defect-engineering-based
implementation for high-performance multilevel data stor-
age in resistive switching memory. IEEE Trans Electron
Dev 60(4):1379–1383. https://doi.org/10.1109/Ted.2013.
2245508
[149] Garbin D, Vianello E, Bichler O, Rafhay Q, Gamrat C,
Ghibaudo G, DeSalvo B, Perniola L (2015) HfO2-based
OxRAM devices as synapses for convolutional neural
networks. IEEE Trans Electron Dev 62(8):2494–2501.
https://doi.org/10.1109/Ted.2015.2440102
[150] Benoist A, Blonkowski S, Jeannot S, Denorme S, Damiens
J, Berger J, Candelier P, Vianello E, Grampeix H, Nodin JF,
Jalaguier E, Perniola L, Allard B (2014) 28 nm Advanced
CMOS resistive RAM solution as embedded non-volatile
memory. In: International reliability physics symposium
[151] Guo T, Tan T, Liu Z (2015) The improved resistive
switching of HfO2: Cu film with multilevel storage. J Mater
Sci 50(21):7043–7047. https://doi.org/10.1007/s10853-
015-9257-9
[152] Wang ZQ, Ambrogio S, Balatti S, Ielmini D (2015) A
2-transistor/1-resistor artificial synapse capable of commu-
nication and stochastic learning in neuromorphic systems.
Front Neurosci-Switz. https://doi.org/10.3389/fnins.2014.
00438
[153] Covi E, Brivio S, Fanciulli M, Spiga S (2015) Synaptic
potentiation and depression in Al:HfO2-based memristor.
Microelectron Eng 147:41–44. https://doi.org/10.1016/j.
mee.2015.04.052
[154] Garbin D, Vianello E, Bichler O, Azzaz M, Rafhay Q,
Candelier P, Gamrat C, Ghibaudo G, DeSalvo B, Pemiola L
(2015) On the impact of OxRAM-based synapses vari-
ability on convolutional neural networks performance. In:
IEEE international symposium nano, pp 193–198
[155] Lee MS, Lee JW, Kim CH, Park BG, Lee JH (2015)
Implementation of short-term plasticity and long-term
potentiation in a synapse using si-based type of charge-trap
memory. IEEE Trans Electron Dev 62(2):569–573. https://
doi.org/10.1109/Ted.2014.2378758
[156] Covi E, Brivio S, Serb A, Prodromakis T, Fanciulli M,
Spiga S (2016) HfO2-based memristors for neuromorphic
applications. IEEE International symposium on circuits
J Mater Sci (2018) 53:8720–8746 8743
Author's personal copy
systems, pp 393–396. https://doi.org/10.1109/iscas.2016.
7527253
[157] Werner T, Garbin D, Vianello E, Bichler O, Cattaert D,
Yvert B, De Salvo B, Perniola L (2016) Real-time decoding
of brain activity by embedded spiking neural networks
using OxRAM synapses. In: IEEE international symposium
on circuits systems, pp 2318–2321
[158] Werner T, Vianello E, Bichler O, Garbin D, Cattaert D,
Yvert B, De Salvo B, Perniola L (2017) Spiking neural
networks based on OxRAM synapses for real-time unsu-
pervised spike sorting (vol 10, pg474, 2016). Front Neu-
rosci-Switz. https://doi.org/10.3389/fnins.2017.00486
[159] Wang IT, Lin YC, Wang YF, Hsu CW, Hou TH (2014) 3D
synaptic architecture with ultralow sub-10 fJ energy per
spike for neuromorphic computation. In: 2014 IEEE inter-
national electron devices meeting (IEDM)
[160] Wang YF, Lin YC, Wang IT, Lin TP, Hou TH (2015)
Characterization and modeling of nonfilamentary Ta/TaOx/
TiO2/Ti analog synaptic device. Sci Rep. https://doi.org/10.
1038/srep10150
[161] Wang IT, Chih-Cheng C, Li-Wen C, Teyuh C, Tuo-Hung H
(2016) 3D Ta/TaO x/TiO 2/Ti synaptic array and linearity
tuning of weight update for hardware neural network
applications. Nanotechnology 27(36):365204
[162] Gao LG, Wang IT, Chen PY, Vrudhula S, Seo JS, Cao Y,
Hou TH, Yu SM (2015) Fully parallel write/read in resistive
synaptic array for accelerating on-chip learning. Nan-
otechnology. https://doi.org/10.1088/0957-4484/26/45/
455204
[163] Wang ZW, Yin MH, Zhang T, Cai YM, Wang YY, Yang
YC, Huang R (2016) Engineering incremental resistive
switching in TaOx based memristors for brain-inspired
computing. Nanoscale 8(29):14015–14022. https://doi.org/
10.1039/c6nr00476h
[164] Wang Q, He DY (2017) Time-decay memristive behavior
and diffusive dynamics in one forget process operated by a
3D vertical Pt/Ta2O5-x/W device. Sci Rep. https://doi.org/
10.1038/s41598-017-00985-0
[165] Yao P, Wu HQ, Gao B, Eryilmaz SB, Huang XY, Zhang
WQ, Zhang QT, Deng N, Shi LP, Wong HSP, Qian H
(2017) Face classification using electronic synapses. Nat
Commun. https://doi.org/10.1038/ncomms15199
[166] Yu SM, Gao B, Fang Z, Yu HY, Kang JF, Wong HSP
(2012) A neuromorphic visual system using RRAM
synaptic devices with sub-pJ energy and tolerance to vari-
ability: experimental characterization and large-scale mod-
eling. In: 2012 IEEE international electron devices meeting
(IEDM)
[167] Berdan R, Vasilaki E, Khiat A, Indiveri G, Serb A, Pro-
dromakis T (2016) Emulating short-term synaptic dynamics
with memristive devices. Sci Rep. https://doi.org/10.1038/
srep18639
[168] Bousoulas P, Asenov P, Karageorgiou I, Sakellaropoulos D,
Stathopoulos S, Tsoukalas D (2016) Engineering amor-
phous-crystalline interfaces in TiO2-x/TiO2-y-based bilayer
structures for enhanced resistive switching and synaptic
properties. J Appl Phys. https://doi.org/10.1063/1.4964872
[169] Mostafa H, Khiat A, Serb A, Mayr CG, Indiveri G, Pro-
dromakis T (2015) Implementation of a spike-based per-
ceptron learning rule using TiO(2 - x) memristors. Front
Neurosci 9:357. https://doi.org/10.3389/fnins.2015.00357
[170] Park J, Kwak M, Moon K, Woo J, Lee D, Hwang H (2016)
TiOx-based RRAM synapse with 64-levels of conductance
and symmetric conductance change by adopting a hybrid
pulse scheme for neuromorphic computing. IEEE Electron
Device Lett 37(12):1559–1562. https://doi.org/10.1109/
Led.2016.2622716
[171] Akoh N, Asai T, Yanagida T, Kawai T, Amemiya Y (2010)
A ReRAM-based analog synaptic device having spike-
timing-dependent plasticity. IEICE Tech Rep
110(246):23–28
[172] Hu SG, Liu Y, Chen TP, Liu Z, Yu Q, Deng LJ, Yin Y,
Hosaka S (2013) Emulating the paired-pulse facilitation of
a biological synapse with a NiOx-based memristor. Appl
Phys Lett. https://doi.org/10.1063/1.4804374
[173] Hu SG, Liu Y, Liu Z, Chen TP, Yu Q, Deng LJ, Yin Y,
Hosaka S (2014) Synaptic long-term potentiation realized
in Pavlov’s dog model based on a NiOx-based memristor.
J Appl Phys. https://doi.org/10.1063/1.4902515
[174] Chang T, Jo SH, Kim KH, Sheridan P, Gaba S, Lu W
(2011) Synaptic behaviors and modeling of a metal oxide
memristive device. Appl Phys A-Mater 102(4):857–863.
https://doi.org/10.1007/s00339-011-6296-1
[175] Fan-Yi M, Shu-Kai D, Li-Dan W, Xiao-Fang H, Zhe-Kang
D (2015) An improved WOx memristor model with
synapse characteristic analysis. Acta Phys Sinica
64(14):148501
[176] Yong Z, Yanling Y, Yuehua P, Weichang Z, Huajun Y,
Zhu’ai Q, Binquan L, Yong Z, Dongsheng T (2014)
Enhanced memristive performance of individual hexagonal
tungsten trioxide nanowires by water adsorption based on
Grotthuss mechanism. Mater Res Express 1(2):025025
[177] Prezioso M, Bayat FM, Hoskins B, Likharev K, Strukov D
(2016) Self-adaptive spike-time-dependent plasticity of
metal-oxide memristors. Sci Rep. https://doi.org/10.1038/
srep21331
[178] Banerjee W, Liu Q, Lv H, Long S, Liu M (2017) Electronic
imitation of behavioral and psychological synaptic activi-
ties using TiOx/Al2O3-based memristor devices. Nanoscale
9(38):14442–14450. https://doi.org/10.1039/C7NR04741J
8744 J Mater Sci (2018) 53:8720–8746
Author's personal copy
[179] Wang CH, He W, Tong Y, Zhao R (2016) Investigation and
manipulation of different analog behaviors of memristor as
electronic synapse for neuromorphic applications. Sci Rep.
https://doi.org/10.1038/srep22970
[180] Kim HJ, Zheng H, Park JS, Kim DH, Kang CJ, Jang JT,
Kim DH, Yoon TS (2017) Artificial synaptic characteristics
with strong analog memristive switching in a Pt/CeO2/Pt
structure. Nanotechnology. https://doi.org/10.1088/1361-
6528/aa712c
[181] Guo LQ, Wan Q, Wan CJ, Zhu LQ, Shi Y (2013) Short-
term memory to long-term memory transition mimicked in
IZO homojunction synaptic transistors. IEEE Electron
Device Lett 34(12):1581–1583. https://doi.org/10.1109/
Led.2013.2286074
[182] Chang YF, Fowler B, Chen YC, Zhou F, Pan CH, Chang
TC, Lee JC (2016) Demonstration of synaptic behaviors
and resistive switching characterizations by proton
exchange reactions in silicon oxide. Sci Rep. https://doi.
org/10.1038/srep21268
[183] Driscoll T, Palit S, Qazilbash MM, Brehm M, Keilmann F,
Chae B-G, Yun S-J, Kim H-T, Cho SY, Jokerst NM, Smith
DR, Basov DN (2008) Dynamic tuning of an infrared
hybrid-metamaterial resonance using vanadium dioxide.
Appl Phys Lett 93(2):024101. https://doi.org/10.1063/1.
2956675
[184] Driscoll T, Kim H-T, Chae B-G, Kim B-J, Lee Y-W, Jokerst
NM, Palit S, Smith DR, Di Ventra M, Basov DN (2009)
memory metamaterials. Science 325(5947):1518–1521.
https://doi.org/10.1126/science.1176580
[185] Lee MJ, Park Y, Suh DS, Lee EH, Seo S, Kim DC, Jung R,
Kang BS, Ahn SE, Lee CB, Seo DH, Cha YK, Yoo IK,
Kim JS, Park BH (2007) Two series oxide resistors appli-
cable to high speed and high density nonvolatile memory.
Adv Mater 19(22):3919–3923. https://doi.org/10.1002/
adma.200700251
[186] Son M, Lee J, Park J, Shin J, Choi G, Jung S, Lee W, Kim
S, Park S, Hwang H (2011) Excellent selector characteris-
tics of nanoscale VO2 for high-density bipolar ReRAM
applications. IEEE Electron Device Lett 32(11):1579–1581.
https://doi.org/10.1109/LED.2011.2163697
[187] Zhang K, Wang B, Wang F, Han Y, Jian X, Zhang H, Wong
HSP (2016) VO2-based selection device for passive resis-
tive random access memory application. IEEE Electron
Device Lett 37(8):978–981. https://doi.org/10.1109/LED.
2016.2582259
[188] Krzysteczko P, Munchenberger J, Schafers M, Reiss G,
Thomas A (2012) The memristive magnetic tunnel junction
as a nanoscopic synapse-neuron system. Adv Mater
24(6):762–766. https://doi.org/10.1002/adma.201103723
[189] Yoshida C, Tsunoda K, Noshiro H, Sugiyama Y (2007)
High speed resistive switching in Pt/TiO2/TiN film for
nonvolatile memory application. Appl Phys Lett
91(22):223510. https://doi.org/10.1063/1.2818691
[190] Yoshida C, Kinoshita K, Yamasaki T, Sugiyama Y (2008)
Direct observation of oxygen movement during resistance
switching in NiO/Pt film. Appl Phys Lett 93(4):042106.
https://doi.org/10.1063/1.2966141
[191] Huang Y-C, Chen P-Y, Chin T-S, Liu R-S, Huang C-Y, Lai
C-H (2012) Improvement of resistive switching in NiO-
based nanowires by inserting Pt layers. Appl Phys Lett
101(15):153106. https://doi.org/10.1063/1.4758482
[192] Chien WC, Lee MH, Feng-Ming L, Lin YY, Hsiang-Lan L,
Hsieh KY, Chih-Yuan L (2011) Multi-level 40 nm WOX
resistive memory with excellent reliability. In: 2011 Inter-
national electron devices meeting, 5–7 Dec. 2011 2011.
pp 313531–313534. https://doi.org/10.1109/iedm.2011.
6131651
[193] Chao X, Biao N, Long Z, Long-Hui Z, Yong-Qiang Y,
Xian-He W, Qun-Ling F, Lin-Bao L, Yu-Cheng W (2013)
High-performance nonvolatile Al/AlO x/CdTe: Sb nano-
wire memory device. Nanotechnology 24(35):355203
[194] Chih Yang Lina CYW, Wua Chung Yi, Hub Chenming,
Tseng Tseung Yuen (2007) Bistable resistive switching in
Al2O3 memory thin films. J Electrochem Soc. https://doi.
org/10.1149/1.2750450
[195] Zhu W, Chen TP, Yang M, Liu Y, Fung S (2012) Resistive
Switching Behavior of Partially Anodized Aluminum Thin
Film at Elevated Temperatures. IEEE Trans Electron
Devices. https://doi.org/10.1109/TED.2012.2205692
[196] Chen YS, Chen B, Gao B, Zhang FF, Qiu YJ, Lian GJ, Liu
LF, Liu XY, Han RQ, Kang JF (2010) Anticrosstalk char-
acteristics correlated with the set process for a-Fe2O3/Nb–
SrTiO3 stack-based resistive switching device. Appl Phys
Lett 97(26):262112. https://doi.org/10.1063/1.3532970
[197] Muraoka S, Osano K, Kanzawa Y, Mitani S, Fujii S,
Katayama K, Katoh Y, Wei Z, Mikawa T, Arita K, Kawa-
shima Y, Azuma R, Kawai K, Shimakawa K, Odagawa A,
Takagi T (2007) Fast switching and long retention Fe-O
ReRAM and its switching mechanism. In: 2007 IEEE
international electron devices meeting, 10–12 Dec. 2007.
pp 779–782. https://doi.org/10.1109/iedm.2007.4419063
[198] Dou C, Kakushima K, Ahmet P, Tsutsui K, Nishiyama A,
Sugii N, Natori K, Hattori T, Iwai H (2012) Resistive
switching behavior of a CeO2 based ReRAM cell incor-
porated with Si buffer layer. Microelectron Reliab
52(4):688–691. https://doi.org/10.1016/j.microrel.2011.10.
019
[199] Lee D, D-j Seong, Jo I, Xiang F, Dong R, Oh S, Hwang H
(2007) Resistance switching of copper doped MoOx films
J Mater Sci (2018) 53:8720–8746 8745
Author's personal copy
for nonvolatile memory applications. Appl Phys Lett
90(12):122104. https://doi.org/10.1063/1.2715002
[200] Li Y, Sinitskii A, Tour JM (2008) Electronic two-terminal
bistable graphitic memories. Nat Mater 7:966. https://doi.
org/10.1038/nmat2331
[201] Syu YE, Zhang R, Chang TC, Tsai TM, Chang KC, Lou
JC, Young TF, Chen JH, Chen MC, Yang YL, Shih CC,
Chu TJ, Chen JY, Pan CH, Su YT, Huang HC, Gan DS, Sze
SM (2013) Endurance improvement technology with
nitrogen implanted in the interface of WSiOx resistance
switching device. IEEE Electron Device Lett
34(7):864–866. https://doi.org/10.1109/LED.2013.2260125
[202] Yu D, Liu LF, Chen B, Zhang FF, Gao B, Fu YH, Liu XY,
Kang JF, Zhang X (2011) Multilevel resistive switching
characteristics in Ag/SiO2/Pt RRAM devices. In: 2011
IEEE international conference of electron devices and
solid-state circuits, 17–18 Nov. 2011. pp 1–2. https://doi.
org/10.1109/edssc.2011.6117721
8746 J Mater Sci (2018) 53:8720–8746
Author's personal copy
![Spiking Neuromorphic Networks with Metal-Oxide …strukov/papers/2016/ISCASspiking2016.pdfmemristors [21-24], our results ar e much less noisy, apparently because we fix the initial](https://img.dokumen.tips/doc/110x75/5aac8ed17f8b9a59658d2aab/spiking-neuromorphic-networks-with-metal-oxide-strukovpapers2016-21-24.jpg)
